chore(release): bump MemHub build number to 13 (#52 )

release: merge MemHub 1.0.4 development fixes (#50 )
* feat(memhub): add model usage audit and cost controls Co-authored-by: Cursor <cursoragent@cursor.com> * fix(memhub): typed source authorization state and managed-key registry ownership Closes #29 Issue #29 surfaced as "WorkBuddy plugin cannot be uninstalled / source sync cannot be turned on", but it is two independent failure domains that were both being reported as a single opaque `desktop_action_failed`: 1. Source sync domain: the UI could not tell whether a source failed to sync because the grant was missing, revoked, or because the source identity or root path changed. Every failure collapsed into one error, so the UI could only say "failed". 2. Host plugin install domain: WorkBuddy shares three registry files (`known_marketplaces.json`, `installed_plugins.json`, `settings.json`) with the host. MemHub recorded a whole-file hash at install time and then refused to uninstall whenever the host (or a reformat) changed any byte anywhere in the file, even in entries MemHub does not own. What changes Domain errors (new `packages/host-bridge/src/domain-error.ts`) - Closed allowlist `DOMAIN_CODES`: grant_required, source_identity_changed, host_uninstall_modified_managed_value, host_uninstall_unknown_owned_file, host_uninstall_needs_review, host_install_journal_invalid, host_install_recovery_required, host_permission_storage_unsafe. - `DomainError` + `domainCodeOf()` so callers stop string-matching on `Error.message`. `diagnosticDomainError()` emits only the code and a repo-relative safe stack, never message text, so diagnostics cannot leak paths or payloads. - Re-exported from `runtime-control` (host-bridge owns the type to avoid a dependency cycle); desktop `observeDesktopAction` now uses `domainCodeOf(error) ?? fallbackCode`. Source authorization state - `SourceAuthorizationState` enum: ok / missing_grant / revoked_grant / source_identity_changed / source_root_unavailable / candidate_changed, plus `SourceGrantStore.authorization_state()` which is read-only and fails closed. - `/api/v1/sources` now returns `authorization_state` per source; contracts and the web app render a concrete next action next to the sync switch instead of a generic failure. - `_change_source_sync` distinguishes sync_prepare_failed, sync_config_write_failed, sync_resume_failed, sync_watcher_failed (409 with the code as detail) while keeping the existing rollback behaviour. - Capture refuses silently but now records a fail-open rejection marker (`sync_disabled` / grant_missing / grant_revoked) that counts occurrences in a time window and carries no payload, path, or session content. WorkBuddy registry ownership (journal v1 -> v2) - Ownership is now the value at MemHub's own key, not the whole file. Added `canonicalJson`, `registryManagedDigest`, `registryWithoutOurs`, `registryHoldsOnlyOurs`. - Install: the plan records `keyPath` + managed digest; apply compares the managed value (CAS) instead of the whole file; the journal writes both the v1 `installed_hash` and the v2 `installed_managed`. - Uninstall: `resolveRegistryOwnership()` compares the v2 digest, falls back to the v1 hash and `isManagedRegistryEntry`, and otherwise produces a review finding instead of throwing. `managedRegistryRemoval()` deletes a file MemHub created when it still holds only our entry; removes just our key when the host has added its own content; and restores recorded bytes only when the rest of the document is unchanged, otherwise patches the current document. - v1 journals migrate: a whole-file hash mismatch no longer blocks uninstall. Tests - `packages/host-bridge/src/domain-error.test.ts` (5): code is carried, message never reaches diagnostics, no guessing on unclassified errors, refuses to parse codes out of free text, allowlist is closed. - `workbuddy-plugin-installer.test.ts` (34 total): uninstall survives a host reformat, removes only our key when the host changed something else, asks for review when a managed entry was changed, migrates a v1 journal, keeps a registry the host now uses, installs over a post-plan change. - `apps/core/tests/test_source_authorization_state.py` (12): every state, fail-closed ordering, capture rejection names, marker contains no secret or path, recording never raises. - `apps/web`: authorization note rendered for revoked grant and hidden when ok. Verification - host-bridge WorkBuddy suite 34/34; domain-error 5/5. - apps/web 1906 tests across 105 files pass. - pnpm -r typecheck clean; biome check clean on touched files. - core-fast: 232 failed / 4131 passed, identical failure count to the clean baseline (232 failed / 4119 passed) with +12 from the new tests; the failures are pre-existing and unrelated (confirmed by stashing and re-running). * feat(memhub): add a start/stop switch for model sedimentation Shipped builds gave the user no way to stop model calls: the model content permission switch is gone (removed with its endpoint in #20) and there is no pause or remove entry either, so a configured model keeps turning conversations into memories and keeps costing money. Adds a user switch in settings: stopping keeps the model configuration and the already imported conversations, leaves queued work pending instead of failing it, and stops new work, enqueueing and backfill so nothing further reaches the model. Starting again resumes the same work without re-entering a key. The flag lives in Core's own user config (config.json), so no schema migration is needed, and the supervisor takes it as an optional callback so every existing caller and test keeps its current behaviour. * test(desktop): keep host-bridge mocks whole and assert the review contract The memhub CI job failed four desktop tests after the issue #29 change. Both were test-side contracts that the new ownership model intentionally moves, not regressions in the fix itself. 1. `host-control-ownership.test.ts` mocked `@one-memory/host-bridge` with a whitelist that carried only `RecallBroker`. Main now imports `domainCodeOf`, which runtime-control re-exports from host-bridge, so the mock handed back undefined and every error path that called it threw a TypeError instead of the error the test was asserting on. The mock now spreads the real module and overrides only `RecallBroker`, which is the part it was written for. 2. `host-permissions.test.ts` asserted the pre-#29 contract, where a change anywhere in a shared registry made uninstall throw `host_uninstall_modified_files`. Ownership is now the value at our own key, so a change to that value is reported instead of guessed at: the plan schedules no removal, names the file in a review finding, and applying it refuses with `host_uninstall_needs_review`. The test keeps the guarantee it existed for, that those bytes are never overwritten, and now also asserts that apply refuses, so the safety is checked at both steps. Verification: full desktop suite 38 files / 436 tests pass, `pnpm -r typecheck` clean, biome reports only two pre-existing warnings on lines this change does not touch. * fix(logging): keep info diagnostics actionable Co-authored-by: Cursor <cursoragent@cursor.com> * test(memhub): register the sedimentation switch test in the core test inventory New core test files must appear in TEST_INDEX.md and in the classification inventory's expected count and digest, and the backfill case needs a settled turn because an unanswered user tail is not backfill material. * feat(diagnostics): expose build info and fullscreen logs Co-authored-by: Cursor <cursoragent@cursor.com> * fix(settings): align help controls and tighten spacing Co-authored-by: Cursor <cursoragent@cursor.com> * release: prepare MemHub 1.0.4 build 8 Co-authored-by: Cursor <cursoragent@cursor.com> * fix(memhub): clear pre-existing dev lint errors and flaky SDK dedup tests Three things blocking CI that came in with dev and are unrelated to the sedimentation switch itself: - biome errors in DebugLogPanel.tsx (import order) and ModelUsageView.tsx (refreshToken is an intentional refresh trigger, annotated like the existing modelVersion trigger in R6Settings). - test_extractor_prompt_override.py's fake LLM now accepts keyword args the real pipeline passes since the usage audit landed (audit_stage). - the fake embeddings in the intention dedup tests used the process-random hash(), so two different texts collided roughly one run in 32 and were misread as duplicates; switched to a deterministic md5 index. * fix(cli): count recall characters as code points A recall whose hits contain an emoji outside the BMP was reported as response_invalid even though the data and the search were fine: the CLI cross-checked the response by summing String.prototype.length, which counts UTF-16 code units, against Core's used_characters, which Python's len() counts in code points. One emoji is two units here and one character there, so any hit set with an emoji failed the check. The sibling validator in host-bridge (context-protocol.ts) already counted code points, so the CLI was the odd one out rather than a deliberate choice. - validateContextResponse now sums through characterCount(), which counts code points, matching Core. - Two regression tests: an emoji hit whose used_characters follows code points must stay grounded, and the same response counted as UTF-16 units must still be refused, so the cross-check is not simply loosened. - The contract now states used_characters counts Unicode code points and is not an encoding length, so a client in another language does not repeat the comparison. Verified by reverting the source change: the first test fails on the old implementation and passes with it. Closes #30 * fix(web): keep the model recovery entry reachable while a probe is pending Closes #40 A pending model probe fenced the configuration editor, and the only recovery entry ("核对测试结果") lived inside the area the fence kept shut. Reaching it required opening the form; opening the form required having recovered. People with a stuck probe could not change or repair their model at all. The fence itself was doing its job and still is. What was wrong is that refusal also hid the way out. R6Settings - openConfiguration() now opens the model area whether or not the editor accepts, so the recovery entry inside it is reachable. editProfile() and add() still return false while the fence is up; nothing about the protection changed. - The refusal notice is reworded to match: the configuration cannot be changed yet, and the next step is down in the area that just opened. Saying "cannot be opened" was both untrue now and unhelpful before. ModelSettings - A refused open is remembered instead of dropped. Once the fence clears, the intent is replayed, so the person lands in the form they asked for. Without this they would get a blank new-configuration form the moment the probe reconciled, which invites a duplicate entry. - Replaying only fills fields in. It never sends a request, so no test is re-sent implicitly; "核对测试结果" remains the only thing that does. Tests - R6Settings: with a current model and without one, the recovery entry is reachable while the editor is fenced, and the refusal is still reported. - ModelSettings: after reconciling, the form reopens on the profile that was asked for, and the probe is sent exactly once. - Both new R6Settings cases and the ModelSettings case fail on the previous code. * perf(runtime-control): decide runtime reuse before staging the payload Every App launch copied the whole 373MB bundle into a temp directory and hashed it twice before asking whether the selected runtime could be reused. On a machine that already runs this build the answer is yes and all of that work is thrown away: measured on a release Mac bundle, 6.9s to copy and about 3.8s of hashing (fileDigests, then verifyInstalledDistribution over the same tree). installPackagedRuntime now describes the payload in place first. The staged tree is a byte for byte copy of resources/{core,host-bridge,runtime,web} plus two generated launchers, so the same relative paths and digests can be read without copying anything. reuseSelectedSameBuild takes that description instead of a staged directory, and only a payload that is going to be installed is ever staged. The decision itself is unchanged: it still compares every file digest except bin/node, still repairs the packaged launcher when the App moved, and still raises installed_version_conflict or distribution_hash_mismatch exactly as before. Nothing was replaced by a cheaper fingerprint, because a wrong reuse is what those errors exist to prevent. Hashing also runs through a small worker pool now; the digests are identical and the manifest keeps its sorted key order. New test: describeBundlePayload must equal the manifest the staged copy produces, so moving the decision ahead of the copy cannot decide something different. Verified read-only against the installed release bundle: 1534 files, identical file sets, zero digest mismatches. * perf(desktop): raise a startup window before the runtime and Core come up Opening the App currently shows nothing at all until initialize finishes. On this machine that is 55.6s from desktop.initializing to desktop.ready (about 19s installing and starting the managed runtime, then about 36s for Core to become ready), and window.requested is logged after desktop.ready. A launch that slow with no window reads as a launch that failed, and people click the icon again. startDesktopLifecycle now takes two optional hooks. showStartupWindow runs before initialize and loads a placeholder; revealStartupWindow runs after initialize and swaps that same window onto the real app. Because the window is reused rather than replaced, there is no second window and no flash. The placeholder is served from the app protocol as /__startup, in memory, not as a file in the web bundle. It performs no IPC: it renders before initialize has registered a single handler, so anything it asked for would fail and leave the page in an error state for the whole wait. That route is also reserved, so a file of the same name inside the bundle can never be reached through it. Raising the placeholder is best effort. It only fills a wait that would otherwise be blank; failing to draw it must not stop the services or the real window. Tests: five lifecycle cases cover the ordering (placeholder before initialize, reveal after, reuse of the same window, nothing for a background launch, and a real window when the placeholder was closed or could not be raised). The four suites that mock ./app-protocol now spread the real module instead of whitelisting it, which is what broke when the protocol gained exports. host-control-ownership gains the BrowserWindow methods the placeholder path uses, and its "no window" assertion now names the placeholder, since a placeholder window is expected there by design. * fix(core): stop the sedimentation switch failing its own request Turning the switch off answered "停止未能确认，模型可能仍在整理记忆，请重试" every time. The value was saved and the error was a lie the endpoint told afterwards: `put_sedimentation` logged `memory_sedimentation.toggled` with `details={"enabled": ...}`, `enabled` is not in the diagnostic allow-list, and `_details` raises. That ValueError escaped the endpoint, so the request is recorded in this machine's log three times as `api.request.failed` / `core.request.unhandled` with `diagnostic_detail_key_invalid`, and the UI took the 500 as an unconfirmed stop. Because the raise sits between the save and the resume, starting again was broken the same way: `resume_allowed()` never ran, so work parked by a stop stayed parked. Three changes, in the order they matter: 1. `enabled` is now a legal detail key. It is a bounded boolean, the same shape as `spawned` and `index_complete`. 2. `diagnostic_event` no longer lets a rejected payload reach the caller. It already swallowed `OSError` with the comment "Diagnostics must not become a new product outage"; an unlisted key is the same failure seen from the caller, and it is exactly what turned every external recall into a 500 and then killed the process last time. The payload is still wrong and still caught in CI — now by every module rather than five. 3. `test_diagnostic_payload_keys.py` walks every module instead of a hand-picked list. The two payloads that caused real outages were in files nobody thought to add to that list; the only shape of the check that holds the line is "every module, always". This file was already failing on dev with `api.py:3673: unlisted details key 'enabled'` — the bug was known to the suite and shipped anyway. `DiagnosticLog.write` still raises, so the rejection itself is unchanged and still tested; only the callers are protected. * fix(core): let the drain claim a manual note's job A memory added by hand stayed on "原话已保存，等待整理" indefinitely, with `task_id: null`: the note's durable job was never claimed, so no task was ever created for it. On this machine the job is still `pending` with `attempts=0` more than eight hours after it was saved, while the write gate admits its source without complaint. The drain had two different answers to one question. `source_writes_permitted` is `_source_writes_on`, which admits `one-memory-manual` by its grant. The claim gate was `_config_store(app).sync_enabled`, the per-app toggle, and `one-memory-manual` is not a connected app: it never appears in the configuration document, so that predicate answers False for it. The job survived the `enabled_sources` prefilter and was then dropped row by row, on every tick, forever. Nothing ever failed and nothing ever ran. The claim gate now asks the same question the write gate asks. Admission is one decision; giving it two predicates is what parked the work. Two tests pin it. One builds the drain as the composition root does and shows the job is claimed, and shows the shipped wiring stranding the same job, so the regression is visible if the toggle comes back. The other records the asymmetry itself: `_source_writes_on` is True for the manual source while `sync_enabled` is False, which is why the pair could disagree at all. No migration and no reinstall: existing pending jobs are picked up on the next drain tick. * fix(memhub): make sedimentation pause stop model dispatch Co-authored-by: Cursor <cursoragent@cursor.com> * feat(memhub): schedule memory writes by event time Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): answer the sedimentation switch before the backlog walk Turning "用模型沉淀记忆" back on wrote the choice to the store and then awaited `resume_allowed()` inside the PUT handler. That call is not a nudge: it runs maintenance over every unfinished task and then a whole-queue scan, and the bounded scan waited -- holding the scan lock -- for the queue to drop under its cohort ceiling. With a real backlog the queue never drops, so the request never returned: one toggle in the field is still open 20+ minutes later with no `api.request.completed`, while the renderer gave up at its 30s timeout and showed "启用未能确认，仍显示上次确认的状态；请重试。" for a switch that had in fact been saved. The choice is the durable fact; the walk is follow-up work: - The handler now stores the choice and returns it, and starts the resume as a watched background task (same idiom as the other Core background walks). Repeated toggles reuse the running walk instead of stacking. - A bounded scan now stops at its ceiling instead of sleeping under the scan lock waiting for capacity, which also un-starves the periodic scan that was queued behind it. It reports `queue_saturated` rather than `no_tasks`, so the status stops claiming there is nothing to do while hundreds wait. Verified on the field log first: the failing toggle is the only request in the window with `api.request.started` and no `api.request.completed`, and its trace shows `memory_sedimentation.toggled` succeeded 6ms in at `pending_count` ~500. New tests fail on the old code (timeout) and pass now. * fix(core): stop a blank author from erasing a whole conversation A user turn whose host did not say who wrote it carries `written_by=unknown`. The write binding's allow-list had every other role/author pair but not that one, so verifying the context raised `hy_memory_write_verification_context_scope` and the task ended as `model_content_outcome_unknown` after two attempts: a memory the model may well have written, that can never be confirmed, and a conversation that is never sedimented again. This is not a rare shape. On this machine 1078 user records across 1002 WorkBuddy conversations are `unknown`, and the failures were 42 per 15 minutes in the field log — every one of them a task reporting `result_count=0` while 402 sat pending. The author is missing, not the turn: the record is still inside the same conversation, workspace and source, which is what the binding actually checks, and the turn assembly already sent it. Accepting it here puts the verification back in agreement with what was sent, the same reasoning that added ("user", AGENT) for sub agent questions. New test fails on the old code at hy_source_context.py:76 — the same frame as the field stack — and passes now. * fix(core): stop an abandoned read reporting a failure nobody owns `in_read_pool` shields the wrapped worker, so a lane that runs out of budget gives up on the answer without yanking a query out from under SQLite. The caller cancels; the read runs on. When such an abandoned read then failed, nothing awaited its future any more: `shield` drops its inner callback as soon as the outer future is cancelled, so the inner future finished with an exception that nobody retrieved. asyncio reports that at GC, and the loop handler files it as `core.loop.task_exception` -- a red error sitting next to a lane that had already degraded on its own budget and already said so. On this machine that is the `raw_context_timeout` pair in the field log: a Future, not a Task, carrying TimeoutError('raw_context_timeout'), reported three times in half an hour. The lane result itself was right; only the diagnostics were arguing with themselves. Retrieving the outcome on completion keeps the diagnostics honest without hiding anything: awaiting the read still raises for a caller who waited, which the second case pins down. The new case fails on the previous code (it reports the TimeoutError) and passes now. * fix(core): one task with no inputs no longer strands every unknown task `MemoryTaskRunner._request` falls back to `chunk[0]` when no record is a person-written user turn. A task whose inputs are gone yields an empty chunk, so that fallback raised IndexError -- an error no caller can read. A reconcile pass catches domain errors per task and moves on, so IndexError did not skip the task: it ended the pass, and every unknown task behind it stayed unknown while the retry backed off 1, 2, 4 ... 128 and gave up. On this machine 68 of 303 unknown tasks had no inputs, which is exactly the `reconcile_unknown / error_type=IndexError` backing off in the field log. Those 68 cannot be reconciled either way -- there is nothing to ask HY about -- but the 235 that can must not be held hostage to them, and this is what stands between a task whose model write may already have landed and its confirmation. An empty chunk is now a domain error, so the pass skips the task and continues. Also folds in one formatting hunk left over in this file from the previous change, so the formatter is clean again. The new case fails on the previous code at memory_processing.py:1036 with the same IndexError the log shows, and passes now. * chore(memhub): bump release build number to 10 The installer refuses a package whose bytes differ while its build number stays the same, so the four Core fixes on dev need their own number to be installable over the 1.0.4 / build 9 package already on the machine. * fix(core): retire a window whose remaining rows FIFO refuses The event-time window asked only whether its ten-minute range still held *some* pending row before deciding the anchor was still active. A pending row that FIFO holds back is not claimable: it may be blocked by an earlier sibling of the same Harness whose own event time sits outside the range, and that sibling can never be claimed from inside this window. The anchor then stayed pinned forever and the whole queue starved. Live shape on one machine: the anchor sat on 2026-10-07T03:24Z while the four pending rows inside it were blocked by 929/930/1243/2347 earlier workbuddy jobs, the earliest of them dated 2026-03-12. Completions stopped at 09:59 and 3.8k queued turns sat at zero output; restarting does not help because the anchor is persisted. The contract (docs/contracts/memory-event-time-scheduling-v1.md) retires a window once every *eligible* pending/running row in the range has settled or parked, and re-anchors on the next oldest eligible job. Reuse the claim's own predicate — free lease, no busy Harness, no earlier unsent sibling — so "the window is busy" and "the window can still be worked" finally agree. Anchor selection and in-window selection already used it; only the liveness check did not. * fix(core): a refused source context is a failure, not an unknown write Issue #44: an OpenCode import carried `role=user, written_by=unknown`. The role/author set now admits it, but the row still never moved: `HyWriteBindings` is built inside `ingest`, and the invocation marker is recorded *before* that call, so the refusal arrived with the marker already set. The classifier checked `hy_invoked` first and parked the task as `unknown` — a state that means "HY may hold a write we cannot confirm" about a request that provably never reached HY. The reason stayed off the row, the task could not be re-driven, and the queue behind it stopped. `hy_memory_write_verification_context_scope` is raised from that construction alone, and `_check_source_context` (which does run after the call) raises `hy_memory_evidence_changed` instead, so the code is unambiguous: nothing was sent. It joins `PRE_SEND_REFUSAL_CODES`, and both classifiers now test that set before the marker: the durable path settles `failed` with the marker cleared so reclaim does not treat a row HY never saw as possibly-sent, and the task path fails with the code instead of marking unknown. Regression covers both directions: a pre-call refusal must end `failed` with `hy_invoked=0`, and a post-call refusal (`authority_invalid`) must still be `unknown` — the fix narrows the unknown set, it does not widen the failure set. * chore(memhub): bump release build number to 11 Carries the window starvation fix (dc0ba49) and the pre-send refusal fix (0df84c0). A new build number keeps the package distinguishable from build 10 and lets it install over the current one. * refactor(memhub): simplify durable memory scheduling Co-authored-by: Cursor <cursoragent@cursor.com> * chore(memhub): bump release build number to 12 Co-authored-by: Cursor <cursoragent@cursor.com> * fix(security): parse diagnostic stack paths linearly Co-authored-by: Cursor <cursoragent@cursor.com> * fix(web): sync sedimentation switch status to home (#51) --------- Co-authored-by: alvinfei <alvinfei@tencent.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: kristinhe <kristinhe@tencent.com>
2026-10-18 18:03:08 +08:00 · 2026-10-18 17:52:59 +08:00 · 2026-10-18 00:37:45 +08:00 · 2026-10-18 00:27:38 +08:00 · 2026-10-17 22:05:24 +08:00 · 2026-10-17 19:40:33 +08:00
482 changed files with 21645 additions and 25615 deletions
@@ -59,4 +59,30 @@ jobs:
      - run: corepack enable
      - run: pnpm install --frozen-lockfile
      - run: uv sync --directory apps/core --all-groups --locked
-      - run: pnpm check
+      - name: Verify public source structure and privacy boundaries
+        run: |
+          python3 scripts/verify_structure.py
+          python3 scripts/generate_core_test_index.py --check
+          python3 scripts/verify_zero_collection.py
+          python3 scripts/verify_zero_collection.py --self-test
+      - name: Lint Node workspaces
+        run: pnpm lint
+      - name: Typecheck Node workspaces
+        run: pnpm typecheck
+      - name: Test Node workspaces
+        run: pnpm test
+      - name: Test release tooling
+        run: pnpm test:release-scripts
+      - name: Fetch and verify the pinned embedding model
+        run: |
+          pnpm fetch:embedding-model
+          uv run --directory apps/core python -c \
+            "from one_memory.embedding_assets import assets_present, verify_assets; assert assets_present(); verify_assets()"
+      - name: Test public Core smoke suite
+        run: >-
+          bash scripts/run-tests-isolated.sh core-pr
+          tests/test_core_test_reporting.py
+          tests/test_import_api.py
+          tests/test_recent_first_import.py
+          tests/test_source_session_count_api.py
+          tests/test_schema_migrations.py
@@ -102,3 +102,65 @@ jobs:
          if-no-files-found: error
          compression-level: 0
          retention-days: 14
+
+  publish-draft-release:
+    if: ${{ inputs.run_lifecycle && startsWith(github.ref, 'refs/tags/v') }}
+    needs: matrix
+    runs-on: ubuntu-24.04
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262
+        with:
+          persist-credentials: false
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093
+        with:
+          pattern: memhub-runtime-linux-*
+          path: memhub/release-assets
+          merge-multiple: true
+      - name: Verify tag and Linux release assets
+        shell: bash
+        run: |
+          set -euo pipefail
+          version=$(node -p 'require("./package.json").version')
+          node scripts/verify-release-version.mjs \
+            --channel stable \
+            --version "$version" \
+            --build-number "$(node -p 'require("./package.json").release.build_number')" \
+            --tag "$GITHUB_REF_NAME"
+          for arch in x64 arm64; do
+            archive="release-assets/memhub-runtime-v${version}-linux-${arch}.tar.gz"
+            test -f "$archive"
+            test -f "$archive.sha256"
+            (
+              cd release-assets
+              sha256sum -c "$(basename "$archive.sha256")"
+            )
+          done
+      - name: Upload Linux assets to a draft GitHub Release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          if gh release view "$GITHUB_REF_NAME" \
+            --repo "$GITHUB_REPOSITORY" >/dev/null 2>&1; then
+            test "$(
+              gh release view "$GITHUB_REF_NAME" \
+                --repo "$GITHUB_REPOSITORY" \
+                --json isDraft \
+                --jq .isDraft
+            )" = "true"
+          else
+            gh release create "$GITHUB_REF_NAME" \
+              --repo "$GITHUB_REPOSITORY" \
+              --verify-tag \
+              --draft \
+              --generate-notes \
+              --title "$GITHUB_REF_NAME"
+          fi
+          gh release upload "$GITHUB_REF_NAME" \
+            release-assets/memhub-runtime-v*-linux-*.tar.gz \
+            release-assets/memhub-runtime-v*-linux-*.tar.gz.sha256 \
+            --repo "$GITHUB_REPOSITORY" \
+            --clobber
@@ -2,6 +2,32 @@

 Hy Memory and MemHub are versioned and released independently.

+## Hy Memory 2.0.3 — 2026-10-17
+
+- Added an optional runtime HTTP transport factory so embedded products can
+  enforce request budgets without wrapping or restricting SDK local-store
+  access.
+- Published
+  [`hy-memory==2.0.3`](https://pypi.org/project/hy-memory/2.0.3/) with
+  unchanged memory extraction, reconciliation, storage, and Coding/Working
+  behavior.
+
+## Unreleased
+
+- Removed MemHub's task-scoped HyPermission store wrappers and global
+  model-content consent switch. The trusted embedded Hy Memory SDK now uses
+  its native local stores directly.
+- Existing installations ignore the retired config flag, readmit previously
+  parked pre-invocation work, and drop only the obsolete
+  `hy_node_permissions` proof table. Evidence, governed memories, model
+  profiles, SDK stores, and source/host grants are preserved.
+
+## Hy Memory 2.0.2 — 2026-10-17
+
+- Published
+  [`hy-memory==2.0.2`](https://pypi.org/project/hy-memory/2.0.2/) with the
+  current SDK fixes and Apache-2.0 package metadata.
+
 ## Hy Memory 2.0.1 — 2026-10-16

 - Published the Python distribution as
@@ -10,13 +36,36 @@ Hy Memory and MemHub are versioned and released independently.
  Because PyPI release metadata is immutable, the next SDK patch release must
  carry the updated license metadata; `2.0.1` retains its published metadata.

-## MemHub
+## MemHub 1.0.3 — Unreleased

-MemHub has no public release yet. The source version is `1.0.0` (build 4), the
-selected npm package is `memhub-cli`, and both npm publication and GitHub
-Release artifacts are still pending. Internal release-candidate builds are not
-public releases.
+- Updated the source version to `1.0.3` (build 7).
+- Pinned the public SDK to `hy-memory==2.0.3`.
+- Removed the HyPermission store wrappers and global model-content consent
+  switch; the trusted embedded SDK now uses native local stores.
+- Prevented uncertain writes from blocking later source work and added one
+  reconciled automatic retry after ten seconds.
+- Added product-facing memory pipeline counters and the log troubleshooting
+  runbook.
+- Allowed a verified packaged Desktop Runtime from a different commit to take
+  over an equal-build install atomically while preserving downgrade and
+  tamper refusal.

-No compatibility or deprecation policy has been adopted beyond the explicit
-CLI/Runtime protocol ranges stored in the release manifests. Future public
-releases must record user-visible changes and supported platforms here.
+## MemHub 1.0.2 — 2026-10-17
+
+- Updated the source version to `1.0.2` (build 6).
+- Pinned the public SDK to `hy-memory==2.0.2`.
+- Pinned `onnxruntime==1.23.2` to retain macOS x64 packaging support.
+- Fixed relaunch and upgrade failures when a prior managed Runtime exists.
+- Embedded canonical build identity when an external packaging pipeline omits
+  release environment fields.
+- Added automatic Linux Runtime upload to tagged Draft Releases.
+
+## MemHub 1.0.1 — Superseded before release
+
+- The candidate tag was not published as a GitHub Release; `1.0.2` replaces it
+  after the managed-Runtime relaunch fix.
+
+## MemHub 1.0.0
+
+- Published the initial GitHub Release with macOS arm64 and Linux Runtime
+  artifacts.
@@ -32,7 +32,7 @@ English | [中文](README.zh.md)

 ## Latest News

-* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **Released** **Hy Memory 2.0 (`hy-memory==2.0.1`)** on [PyPI](https://pypi.org/project/hy-memory/2.0.1/) and open-sourced the **Hy-Memory** monorepo together with **[MemHub](memhub/README.md)**!
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **Released** **Hy Memory 2.0 (`hy-memory==2.0.3`)** on [PyPI](https://pypi.org/project/hy-memory/2.0.3/) and open-sourced the **Hy-Memory** monorepo together with **[MemHub](memhub/README.md)**!
 * *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **Open-sourced** **[Hy-Memory-30B-A3B](https://huggingface.co/tencent/Hy-Memory-30B-A3B)** on Hugging Face along with its [technical report](https://arxiv.org/abs/2610.22196) on attributed reinforcement learning for long-term memory writing!
 * *<mark>[new]</mark>* 🔥🔥 **[2026-09]**: 📊 **Released** **[CMB: A Coding Memory Benchmark for Long-Horizon and Interleaved Coding Agents](docs/papers/CMB_A_Coding_Memory_Benchmark_for_Long_Horizon_and_Interleaved_Coding_Agents.pdf)** (submitted to **ICLR 2027**).
 * *<mark>[new]</mark>* 🔥🔥 **[2026-08]**: 🎉 **Accepted** at **EMNLP 2026**: **[Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents](https://arxiv.org/abs/2606.09483)**!
@@ -63,13 +63,13 @@ Welcome to the **Hy Memory Research Family**, covering cognitive chat memory, lo
 ### 1. Installation

 ```bash
-pip install hy-memory==2.0.1              # core (includes ChromaDB and Kuzu)
-pip install "hy-memory[qdrant]==2.0.1"    # + Qdrant
-pip install "hy-memory[faiss]==2.0.1"     # + FAISS
-pip install "hy-memory[graph]==2.0.1"     # + Neo4j
-pip install "hy-memory[mysql]==2.0.1"     # + MySQL cache
-pip install "hy-memory[rerank]==2.0.1"    # + local multilingual cross-encoder rerank
-pip install "hy-memory[all]==2.0.1"       # everything
+pip install hy-memory==2.0.3              # core (includes ChromaDB and Kuzu)
+pip install "hy-memory[qdrant]==2.0.3"    # + Qdrant
+pip install "hy-memory[faiss]==2.0.3"     # + FAISS
+pip install "hy-memory[graph]==2.0.3"     # + Neo4j
+pip install "hy-memory[mysql]==2.0.3"     # + MySQL cache
+pip install "hy-memory[rerank]==2.0.3"    # + local multilingual cross-encoder rerank
+pip install "hy-memory[all]==2.0.3"       # everything
 ```

 ### 2. Configuration
@@ -32,7 +32,7 @@

 ## 最新动态

-* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **正式发布** **Hy Memory 2.0 (`hy-memory==2.0.1`)**（已上线 [PyPI](https://pypi.org/project/hy-memory/2.0.1/)），并开源 **Hy-Memory** monorepo 与 **[MemHub](memhub/README.md)**！
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **正式发布** **Hy Memory 2.0 (`hy-memory==2.0.3`)**（已上线 [PyPI](https://pypi.org/project/hy-memory/2.0.3/)），并开源 **Hy-Memory** monorepo 与 **[MemHub](memhub/README.md)**！
 * *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **模型开源** 在 Hugging Face 发布 **[Hy-Memory-30B-A3B](https://huggingface.co/tencent/Hy-Memory-30B-A3B)** 长期记忆写入模型及[技术报告](https://arxiv.org/abs/2610.22196)！
 * *<mark>[new]</mark>* 🔥🔥 **[2026-09]**: 📊 **基准与方法发布** 推出长周期与交错式编程记忆评测基准 **[CMB: A Coding Memory Benchmark for Long-Horizon and Interleaved Coding Agents](docs/papers/CMB_A_Coding_Memory_Benchmark_for_Long_Horizon_and_Interleaved_Coding_Agents.pdf)**（已投稿 **ICLR 2027**）。
 * *<mark>[new]</mark>* 🔥🔥 **[2026-08]**: 🎉 **论文录用** 双系统认知记忆架构论文 **[Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents](https://arxiv.org/abs/2606.09483)** 被 **EMNLP 2026** 接收！
@@ -63,13 +63,13 @@
 ### 1. 安装

 ```bash
-pip install hy-memory==2.0.1              # 核心（含 ChromaDB 与 Kuzu）
-pip install "hy-memory[qdrant]==2.0.1"    # + Qdrant
-pip install "hy-memory[faiss]==2.0.1"     # + FAISS
-pip install "hy-memory[graph]==2.0.1"     # + Neo4j
-pip install "hy-memory[mysql]==2.0.1"     # + MySQL 缓存
-pip install "hy-memory[rerank]==2.0.1"    # + 本地多语言 cross-encoder rerank
-pip install "hy-memory[all]==2.0.1"       # 全部
+pip install hy-memory==2.0.3              # 核心（含 ChromaDB 与 Kuzu）
+pip install "hy-memory[qdrant]==2.0.3"    # + Qdrant
+pip install "hy-memory[faiss]==2.0.3"     # + FAISS
+pip install "hy-memory[graph]==2.0.3"     # + Neo4j
+pip install "hy-memory[mysql]==2.0.3"     # + MySQL 缓存
+pip install "hy-memory[rerank]==2.0.3"    # + 本地多语言 cross-encoder rerank
+pip install "hy-memory[all]==2.0.3"       # 全部
 ```

 ### 2. 配置
@@ -37,7 +37,7 @@ do not belong in the public source tree.
 - Root GitHub workflows run SDK and MemHub checks from their respective working
  directories. Native lifecycle workflows produce reviewable Actions artifacts;
  tag releases remain fail-closed on public identity, signing and notarization.
- Hy Memory is published as `hy-memory==2.0.1`. MemHub pins that public
+- Hy Memory is published as `hy-memory==2.0.3`. MemHub pins that public
  distribution; its selected npm package is `memhub-cli`, which is not yet
  published.
 - Hy-Memory source uses Apache-2.0 at the repository root. Third-party
@@ -5,7 +5,7 @@
 ## 安装与快速开始

 ```bash
-pip install hy-memory==2.0.1
+pip install hy-memory==2.0.3

 # 可选后端
 pip install 'hy-memory[qdrant]==2.0.1'
@@ -1123,6 +1123,7 @@ class Extractor:
                max_tokens=self._llm_config.agent_max_tokens,
                tools=None,
                tool_choice=None,
+                audit_stage="extract",
            )
            self._call_count += 1
            total_tokens += response.tokens_used
@@ -1233,6 +1234,7 @@ class Extractor:
                        max_tokens=self._llm_config.agent_max_tokens,
                        tools=None,
                        tool_choice=None,
+                        audit_stage="extract",
                    )
                    self._call_count += 1
                    total_tokens += response.tokens_used
@@ -34,6 +34,7 @@ import os
 import weakref

 from ..config import MemoryConfig
+from ..utils.model_call_context import model_call_stage
 from ..core.workload import llm_slot

 logger = logging.getLogger(__name__)
@@ -348,6 +349,7 @@ class LLMProvider:
        stop: List[str] = None,
        tools: Optional[List[Dict[str, Any]]] = None,
        tool_choice: Optional[Any] = None,
+        audit_stage: Optional[str] = None,
        **kwargs
    ) -> LLMResponse:
        """
@@ -386,15 +388,16 @@ class LLMProvider:
                        len(prompt), _content_fingerprint(prompt),
                    )

-                response = await self._call_llm(
-                    prompt=prompt,
-                    max_tokens=max_tokens,
-                    temperature=temperature,
-                    stop=stop,
-                    tools=tools,
-                    tool_choice=tool_choice,
-                    **kwargs
-                )
+                with model_call_stage(audit_stage):
+                    response = await self._call_llm(
+                        prompt=prompt,
+                        max_tokens=max_tokens,
+                        temperature=temperature,
+                        stop=stop,
+                        tools=tools,
+                        tool_choice=tool_choice,
+                        **kwargs
+                    )

                self._total_calls += 1
                self._total_tokens += response.tokens_used
@@ -496,11 +499,21 @@ class LLMProvider:
        clients = self._clients.setdefault(loop, {})
        client = clients.get(client_key)
        if client is None:
+            client_options: Dict[str, Any] = {
+                "api_key": base_api_key,
+                "base_url": base_url,
+                "timeout": self._llm_config.timeout,
+                "max_retries": 0,
+            }
+            transport_factory = getattr(self.config, "http_transport_factory", None)
+            if callable(transport_factory):
+                import httpx
+
+                client_options["http_client"] = httpx.AsyncClient(
+                    transport=transport_factory()
+                )
            client = AsyncOpenAI(
-                api_key=base_api_key,
-                base_url=base_url,
-                timeout=self._llm_config.timeout,
-                max_retries=0,
+                **client_options,
            )
            clients[client_key] = client

@@ -585,6 +598,7 @@ class LLMProvider:
        temperature: Optional[float] = None,
        tools: Optional[List[Dict[str, Any]]] = None,
        tool_choice: Optional[Any] = None,
+        audit_stage: Optional[str] = None,
    ) -> LLMResponse:
        """
        多轮 messages 调用（支持 tool results 回传）。
@@ -612,13 +626,14 @@ class LLMProvider:
                        f"msgs={len(messages)} tools={len(tools) if tools else 0}"
                    )

-                response = await self._chat_completions(
-                    messages,
-                    max_tokens=max_tokens,
-                    temperature=temperature,
-                    tools=tools,
-                    tool_choice=tool_choice,
-                )
+                with model_call_stage(audit_stage):
+                    response = await self._chat_completions(
+                        messages,
+                        max_tokens=max_tokens,
+                        temperature=temperature,
+                        tools=tools,
+                        tool_choice=tool_choice,
+                    )

                self._total_calls += 1
                self._total_tokens += response.tokens_used
@@ -669,6 +669,7 @@ class MemoryReconciler:
            response = await llm.complete(
                prompt=prompt,
                max_tokens=500,
+                audit_stage="reconcile",
            )
            text = response.content.strip()
            if "```json" in text:
@@ -1383,6 +1384,7 @@ class MemoryReconciler:
                    response = await llm.complete(
                        prompt=prompt,
                        max_tokens=dyn_max_tokens,
+                        audit_stage="reconcile",
                    )
                except Exception as e:
                    # LLM 彻底不可用（provider 已按 max_retries 重试过，网关 500 /
@@ -919,6 +919,7 @@ class CodingMemoryExtractor:
            resp = await self.llm_provider.complete(
                prompt=prompt,
                max_tokens=self.max_tokens,
+                audit_stage="coding_extract",
            )
        except Exception as e:
            logger.warning(f"[coding-extract] LLM call failed: {e}")
@@ -571,6 +571,7 @@ class WorkingFactReconciler:
        resp = await self.llm_provider.complete(
            prompt=prompt,
            max_tokens=self.max_tokens,
+            audit_stage="coding_fact_reconcile",
        )
        return self._parse_ops(resp.content)

@@ -637,6 +638,7 @@ class WorkingFactReconciler:
            resp = await self.llm_provider.complete(
                prompt=prompt,
                max_tokens=_CHAIN_SUMMARY_MAX_TOKENS,
+                audit_stage="coding_fact_summary",
            )
        except Exception as e:
            logger.warning(f"[fact-reconcile] chain summary failed: {e}")
@@ -29,6 +29,9 @@ logger = logging.getLogger(__name__)

 _JUDGE_MAX_TURNS = 20
 _JUDGE_HEAD_TURNS = 5
+_JUDGE_USER_CONTENT_LIMIT = 1_000
+_JUDGE_USER_CONTENT_HEAD = 200
+_JUDGE_USER_CONTENT_TAIL = 200
 _JUDGE_MAX_QUERY_CHARS = 300
 _JUDGE_MAX_QUERIES = 20

@@ -186,6 +189,7 @@ async def classify_messages_scene(
            prompt=prompt,
            max_tokens=max_tokens,
            temperature=temperature,
+            audit_stage="coding_judge",
        )
        _elapsed = round((time.perf_counter() - _t0) * 1000, 2)
        result = _parse_scene_json(resp.content)
@@ -631,9 +635,13 @@ def _format_turns(summary: List[Dict[str, Any]]) -> str:
    rows = []
    for t in summary:
        user = t.get("user", "") or ""
-        # 单 turn 用户文本截到 200 字符
-        if len(user) > 200:
-            user = user[:200] + "..."
+        if len(user) > _JUDGE_USER_CONTENT_LIMIT:
+            omitted = len(user) - _JUDGE_USER_CONTENT_HEAD - _JUDGE_USER_CONTENT_TAIL
+            user = (
+                user[:_JUDGE_USER_CONTENT_HEAD]
+                + f"\n[...{omitted} chars omitted...]\n"
+                + user[-_JUDGE_USER_CONTENT_TAIL:]
+            )
        tools = t.get("tools", []) or []
        rows.append(
            f"- turn {t.get('turn', '?')}: user={user!r}, tools={tools}"
@@ -310,6 +310,7 @@ class CodingMemoryReconciler:
        resp = await self.llm_provider.complete(
            prompt=prompt,
            max_tokens=self.max_tokens,
+            audit_stage="coding_reconcile",
        )
        return self._parse_ops(resp.content), prompt, resp.content, resp

@@ -6,7 +6,7 @@ Agent Memory - 配置管理

 import os
 from dataclasses import dataclass, field
-from typing import Optional, Dict, Any, List
+from typing import Optional, Dict, Any, Callable, List
 from pathlib import Path


@@ -1252,6 +1252,13 @@ class MemoryConfig:
    enable_agent: bool = False
    debug: bool = False
    metrics_enabled: bool = True
+    # Runtime-only HTTP transport factory. MemHub uses this for request-cost
+    # budgeting without restricting the embedded SDK's local store access.
+    http_transport_factory: Optional[Callable[[], Any]] = field(
+        default=None,
+        repr=False,
+        compare=False,
+    )

    def __post_init__(self):
        # enable_graph 跟随 mode 走：mode=ultra 时自动启用，无需独立环境变量
@@ -356,11 +356,21 @@ class EmbedService:
        key = self._client_cache_key(base_url, api_key, timeout)
        client = self._client_cache.get(key)
        if client is None:
+            client_options = {
+                "api_key": api_key,
+                "base_url": base_url or None,
+                "timeout": timeout,
+                "max_retries": 0,
+            }
+            transport_factory = getattr(self.config, "http_transport_factory", None)
+            if callable(transport_factory):
+                import httpx
+
+                client_options["http_client"] = httpx.AsyncClient(
+                    transport=transport_factory()
+                )
            client = AsyncOpenAI(
-                api_key=api_key,
-                base_url=base_url or None,
-                timeout=timeout,
-                max_retries=0,
+                **client_options,
            )
            self._client_cache[key] = client
        return client
@@ -0,0 +1,24 @@
+"""Context propagated to host-provided HTTP transports for cost accounting."""
+
+from collections.abc import Iterator
+from contextlib import contextmanager
+from contextvars import ContextVar, Token
+
+_model_call_stage: ContextVar[str] = ContextVar(
+    "hy_memory_model_call_stage",
+    default="other",
+)
+
+
+def current_model_call_stage() -> str:
+    return _model_call_stage.get()
+
+
+@contextmanager
+def model_call_stage(stage: str | None) -> Iterator[None]:
+    normalized = (stage or "other").strip().lower() or "other"
+    token: Token[str] = _model_call_stage.set(normalized)
+    try:
+        yield
+    finally:
+        _model_call_stage.reset(token)
@@ -80,7 +80,11 @@ class PipelineLogWriter:
                split_namespace_user,
            )

-            tokens = (prompt_tokens or 0) + (completion_tokens or 0) + (total_tokens or 0)
+            # Providers define total_tokens as prompt+completion. Adding all
+            # three fields doubled every cost estimate in this JSONL.
+            tokens = (total_tokens or 0) or (
+                (prompt_tokens or 0) + (completion_tokens or 0)
+            )
            namespace, uid = split_namespace_user(user_id or "")
            if not is_llm_content_logging_enabled():
                prompt = redact_content(prompt)
@@ -15,12 +15,12 @@ write、search，并把搜索结果格式化为可交给模型的 memory text ch

 ```text
 distribution: hy-memory
-version:      2.0.1
+version:      2.0.3
 mode:         按运行时配置选择
 reader:       hybrid
 ```

-仓库 pin、lockfile 和运行时版本守卫均固定到公开的 `hy-memory==2.0.1`。
+仓库 pin、lockfile 和运行时版本守卫均固定到公开的 `hy-memory==2.0.3`。
 MemHub 在本地 Runtime 内嵌入 SDK，但不绕过自身的来源授权、模型授权、Evidence、
 Governance 和撤权边界。

@@ -37,6 +37,40 @@ MemHub 对 SDK 使用以下公开、可复现的本地策略：
 - API Key 进入系统安全存储，不写入仓库、安装包、日志或 HY metadata；
 - 服务端数据库、租户队列及宿主 shell 中的 ambient 凭据不会进入 SDK config。

+发布包固定应用以下 SDK 行为配置：
+
+| 环境变量 | 值 |
+|---|---|
+| `MEMORY_MODE` | `ultra` |
+| `MEMORY_ENABLE_GRAPH` | `true` |
+| `MEMORY_GRAPH_RERANK_ENABLED` | `false` |
+| `MEMORY_GRAPH_TEMPORAL_SPINE` | `true` |
+| `MEMORY_TOPIC_SUMMARY_ENABLED` | `false` |
+| `MEMORY_EPISODE_SUMMARY_ENABLED` | `false` |
+| `MEMORY_EPISODE_RECALL_ENABLED` | `true` |
+| `MEMORY_DYNAMIC_EPISODE_ENABLED` | `true` |
+| `MEMORY_SESSION_EPISODE_ENABLED` | `false` |
+| `MEMORY_SESSION_EPISODE_CONTEXT_ENABLED` | `false` |
+| `MEMORY_EPISODE_CAPSULE_CONTEXT_ENABLED` | `true` |
+| `MEMORY_EPISODE_CAPSULE_QUERY_MIN` | `0.65` |
+| `MEMORY_EPISODE_CAPSULE_ASSIGNMENT_MIN` | `0.60` |
+| `MEMORY_EPISODE_CAPSULE_FIRST_RANK_MAX` | `3` |
+| `MEMORY_EPISODE_CAPSULE_MEMBER_LIMIT` | `3` |
+| `MEMORY_ENABLE_INTENTION` | `false` |
+| `MEMORY_ENABLE_QUERY_REWRITE` | `false` |
+| `MEMORY_FEW_SHOT_ENABLED` | `true` |
+| `MEMORY_HYBRID_V2_QUERY_EXPAND` | `1` |
+| `MEMORY_HYBRID_V2_GEO_INTENT_PRIOR` | `0.6` |
+| `MEMORY_HYBRID_V2_DIVERSITY` | `1` |
+| `MEMORY_HYBRID_V2_TEMPORAL_ADMIT_PRIOR` | `0.55` |
+| `MEMORY_HYBRID_V2_TEMPORAL_BOOST` | `0.18` |
+| `MEMORY_CROSS_DOMAIN_CON_THRESHOLD` | `0.5` |
+| `MEMORY_READER_ENABLE_SUMMARY` | `false` |
+| `MEMORY_SUMMARY_ENABLED_IN_SYS2` | `false` |
+| `MEMORY_LOCAL_RERANK_ENABLED` | `false` |
+| `RECONCILE_ENABLED` | `true` |
+| `HY_MEMORY_READER` | `hybrid` |
+
 这些默认值集中在
 `apps/core/src/one_memory/hy_runtime_config.py`，并由发布测试和 lockfile 共同约束。
 独立脚本若绕过 MemHub 直接导入 SDK，应自行显式配置存储目录和模型端点。
@@ -271,9 +305,9 @@ Runtime 应长期复用 client，不为每次请求重新初始化数据库和

 ```text
 Evidence
-  -> ModelProcessingPermissionLease
-  -> HyPermissionTransport
-  -> HyMemoryBridge.ingest / ingest_extracted / search_ids
+  -> HyMemoryBridge
+  -> embedded Hy Memory SDK
+  -> native Vector / Coding / Graph stores
  -> write result verification
  -> Governance
  -> Recall Broker / UI
@@ -282,8 +316,11 @@ Evidence
 不要在新业务模块里直接创建 `HyMemoryClient`，否则会绕过：

 - 来源读取授权；
- 模型内容处理授权；
- 发送前与响应后的撤权复核；
+- 共享 Runtime、模型配置、API Key 安全存储与请求预算；
 - Evidence 引用与作者归属；
 - 写后回读验证；
 - Governance 状态、版本、更正和删除级联。
+
+Hy Memory SDK 是 MemHub 的内置可信组件，可自由访问和管理本机的全部
+Hy Memory 数据。来源授权仍决定哪些宿主数据可以进入本机；宿主 Recall Grant
+仍决定哪些外部 AI 应用可以读取治理后的记忆。这两类边界不限制内置 SDK 的本地 CRUD。
@@ -1,18 +1,30 @@
-# MemHub / One Memory
+# MemHub

-MemHub 是一个本地优先的共享记忆产品：把用户授权的多个 AI 产品中的历史对话和新增内容
-汇总到同一份记忆，并让这些 AI 在独立授权后复用它。Hy Memory 是 MemHub 使用的 Python
-SDK 和记忆引擎。One Memory / `one-memory` 是历史模块、数据目录、服务和兼容命令名称。
+MemHub是一个本地优先的共享记忆产品。它把用户明确授权的AI对话和编码记录汇总到一份本地记忆中，再由用户分别授权给支持的AI工具调用。你可以继续使用熟悉的客户端，同时保留跨工具延续的上下文。Hy Memory是MemHub使用的Python SDK和记忆引擎。
+
+# 💡❤️ MemHub 刚刚开源，欢迎一起优化共建！ ❤️💡 
+
+**# 💭 MemHub处于preview阶段，当前公开版本可以下载和体验，部分平台制品、安装路径与宿主体验仍在持续优化验证。安装前请先查看Release里的实际制品和版本说明。**
+
+> 🙋 如果你遇到问题/有建议：欢迎提交Issue，并附上MemHub版本、操作系统与芯片架构、复现步骤、预期结果、实际结果和经过脱敏的日志。请不要上传APIKey、真实对话、个人路径或其他敏感信息。
+
+> 我们会公开跟进可复现的问题，持续发布修复，并根据真实反馈完善安装、同步和召回体验。
+
+**# 感谢每一条反馈，期待你和我们一起打磨MemHub的体验 ：）**
+
+> 💗 MemHub has just been open-sourced and is under active development. The current public release is available for download andevaluation. Some platform packages, installation paths, and host integrations are still being completed and verified. Check the assets
+and release notes before installing.
+If you run into a problem, please open an Issue with the version, platform,reproduction steps, and sanitized logs, We wilreviewreports, publish fixes, and keep the support matrix up to date.

 ## 发行与平台合同

-公共 Python SDK 为 `hy-memory==2.0.1`。MemHub 源码版本为 `1.0.0`
-（build 4），GitHub 仓库为
+公共 Python SDK 为 `hy-memory==2.0.3`。MemHub 源码版本为 `1.0.3`
+（build 7），GitHub 仓库为
 [`Tencent-Hunyuan/Hy-Memory`](https://github.com/Tencent-Hunyuan/Hy-Memory)，npm 包名为
 `memhub-cli`。当前代码架构见
 [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md)。

-`v1.0.0` 的发行目标是 Linux x64/arm64 Runtime 与 macOS arm64/x64 App 和 Runtime。
+`v1.0.3` 的发行目标是 Linux x64/arm64 Runtime 与 macOS arm64/x64 App 和 Runtime。
 Linux 制品由 GitHub Actions 在原生架构 runner 上构建并完成生命周期验收；macOS
 制品由发行流水线完成 Developer ID 签名、公证和验收后手动上传。Windows 不支持。
 只有同时出现在 GitHub Release 与 `release-manifest.json` 中、且通过公开校验的制品才属于
@@ -42,6 +54,8 @@ Linux 制品由 GitHub Actions 在原生架构 runner 上构建并完成生命

 架构详情见 [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) 和
 [`docs/contracts/runtime-unification.md`](docs/contracts/runtime-unification.md)。
+产品、测试和一线支持排查导入与记忆沉淀时，使用
+[`docs/PRODUCT_LOG_TROUBLESHOOTING.md`](docs/PRODUCT_LOG_TROUBLESHOOTING.md)。

 ## 开发环境

@@ -99,7 +113,7 @@ MemHub 的 SQLite 数据默认位于 `~/.one-memory`，不在 npm 包或 Runtime

 ## Hy Memory

-当前开发依赖固定为公开 PyPI 包 `hy-memory==2.0.1`。SDK 配置、权限边界及 search
+当前开发依赖固定为公开 PyPI 包 `hy-memory==2.0.3`。SDK 配置、权限边界及 search
 返回格式见 [`MEMORY.md`](MEMORY.md)。HY search 返回的 `content` 已包含 SDK 统一生成的
 逻辑时间和演化上下文；调用方不应再次转换时间戳或重复拼接 `When`。

@@ -3,9 +3,9 @@
  "dataLicense": "CC0-1.0",
  "SPDXID": "SPDXRef-DOCUMENT",
  "name": "one-memory",
-  "documentNamespace": "https://one-memory.local/spdx/2026-10-15T17:26:47Z",
+  "documentNamespace": "https://one-memory.local/spdx/2026-10-17T13:52:48Z",
  "creationInfo": {
-    "created": "2026-10-15T17:26:47Z",
+    "created": "2026-10-17T13:52:48Z",
    "creators": [
      "Tool: one-memory-generate-sbom"
    ]
@@ -479,6 +479,23 @@
        }
      ]
    },
+    {
+      "SPDXID": "SPDXRef-Package-pypi-coloredlogs",
+      "name": "coloredlogs",
+      "versionInfo": "15.0.1",
+      "downloadLocation": "NOASSERTION",
+      "filesAnalyzed": false,
+      "licenseConcluded": "NOASSERTION",
+      "licenseDeclared": "NOASSERTION",
+      "copyrightText": "NOASSERTION",
+      "externalRefs": [
+        {
+          "referenceCategory": "PACKAGE-MANAGER",
+          "referenceType": "purl",
+          "referenceLocator": "pkg:pypi/coloredlogs@15.0.1"
+        }
+      ]
+    },
    {
      "SPDXID": "SPDXRef-Package-pypi-distro",
      "name": "distro",
@@ -751,6 +768,23 @@
        }
      ]
    },
+    {
+      "SPDXID": "SPDXRef-Package-pypi-humanfriendly",
+      "name": "humanfriendly",
+      "versionInfo": "10.0",
+      "downloadLocation": "NOASSERTION",
+      "filesAnalyzed": false,
+      "licenseConcluded": "NOASSERTION",
+      "licenseDeclared": "NOASSERTION",
+      "copyrightText": "NOASSERTION",
+      "externalRefs": [
+        {
+          "referenceCategory": "PACKAGE-MANAGER",
+          "referenceType": "purl",
+          "referenceLocator": "pkg:pypi/humanfriendly@10.0"
+        }
+      ]
+    },
    {
      "SPDXID": "SPDXRef-Package-pypi-idna",
      "name": "idna",
@@ -1125,6 +1159,23 @@
        }
      ]
    },
+    {
+      "SPDXID": "SPDXRef-Package-pypi-mpmath",
+      "name": "mpmath",
+      "versionInfo": "1.3.0",
+      "downloadLocation": "NOASSERTION",
+      "filesAnalyzed": false,
+      "licenseConcluded": "NOASSERTION",
+      "licenseDeclared": "NOASSERTION",
+      "copyrightText": "NOASSERTION",
+      "externalRefs": [
+        {
+          "referenceCategory": "PACKAGE-MANAGER",
+          "referenceType": "purl",
+          "referenceLocator": "pkg:pypi/mpmath@1.3.0"
+        }
+      ]
+    },
    {
      "SPDXID": "SPDXRef-Package-pypi-multidict",
      "name": "multidict",
@@ -1230,7 +1281,7 @@
    {
      "SPDXID": "SPDXRef-Package-pypi-onnxruntime",
      "name": "onnxruntime",
-      "versionInfo": "1.29.0",
+      "versionInfo": "1.23.2",
      "downloadLocation": "NOASSERTION",
      "filesAnalyzed": false,
      "licenseConcluded": "NOASSERTION",
@@ -1240,7 +1291,7 @@
        {
          "referenceCategory": "PACKAGE-MANAGER",
          "referenceType": "purl",
-          "referenceLocator": "pkg:pypi/onnxruntime@1.29.0"
+          "referenceLocator": "pkg:pypi/onnxruntime@1.23.2"
        }
      ]
    },
@@ -1992,6 +2043,23 @@
        }
      ]
    },
+    {
+      "SPDXID": "SPDXRef-Package-pypi-sympy",
+      "name": "sympy",
+      "versionInfo": "1.14.0",
+      "downloadLocation": "NOASSERTION",
+      "filesAnalyzed": false,
+      "licenseConcluded": "NOASSERTION",
+      "licenseDeclared": "NOASSERTION",
+      "copyrightText": "NOASSERTION",
+      "externalRefs": [
+        {
+          "referenceCategory": "PACKAGE-MANAGER",
+          "referenceType": "purl",
+          "referenceLocator": "pkg:pypi/sympy@1.14.0"
+        }
+      ]
+    },
    {
      "SPDXID": "SPDXRef-Package-pypi-tenacity",
      "name": "tenacity",
@@ -2267,12 +2335,12 @@
    {
      "SPDXID": "SPDXRef-Package-hy-memory",
      "name": "hy-memory",
-      "versionInfo": "2.0.1",
+      "versionInfo": "2.0.4",
      "downloadLocation": "NOASSERTION",
      "filesAnalyzed": false,
-      "licenseConcluded": "MIT",
-      "licenseDeclared": "MIT",
-      "copyrightText": "Copyright (c) 2025 HY Memory Contributors"
+      "licenseConcluded": "Apache-2.0",
+      "licenseDeclared": "Apache-2.0",
+      "copyrightText": "Copyright (C) 2026 Tencent. All rights reserved."
    },
    {
      "SPDXID": "SPDXRef-Package-memmy-agent-derived",
@@ -2427,6 +2495,11 @@
      "relationshipType": "DEPENDS_ON",
      "relatedSpdxElement": "SPDXRef-Package-pypi-cloudpickle"
    },
+    {
+      "spdxElementId": "SPDXRef-Package-one-memory",
+      "relationshipType": "DEPENDS_ON",
+      "relatedSpdxElement": "SPDXRef-Package-pypi-coloredlogs"
+    },
    {
      "spdxElementId": "SPDXRef-Package-one-memory",
      "relationshipType": "DEPENDS_ON",
@@ -2507,6 +2580,11 @@
      "relationshipType": "DEPENDS_ON",
      "relatedSpdxElement": "SPDXRef-Package-pypi-huggingface-hub"
    },
+    {
+      "spdxElementId": "SPDXRef-Package-one-memory",
+      "relationshipType": "DEPENDS_ON",
+      "relatedSpdxElement": "SPDXRef-Package-pypi-humanfriendly"
+    },
    {
      "spdxElementId": "SPDXRef-Package-one-memory",
      "relationshipType": "DEPENDS_ON",
@@ -2617,6 +2695,11 @@
      "relationshipType": "DEPENDS_ON",
      "relatedSpdxElement": "SPDXRef-Package-pypi-more-itertools"
    },
+    {
+      "spdxElementId": "SPDXRef-Package-one-memory",
+      "relationshipType": "DEPENDS_ON",
+      "relatedSpdxElement": "SPDXRef-Package-pypi-mpmath"
+    },
    {
      "spdxElementId": "SPDXRef-Package-one-memory",
      "relationshipType": "DEPENDS_ON",
@@ -2872,6 +2955,11 @@
      "relationshipType": "DEPENDS_ON",
      "relatedSpdxElement": "SPDXRef-Package-pypi-starlette"
    },
+    {
+      "spdxElementId": "SPDXRef-Package-one-memory",
+      "relationshipType": "DEPENDS_ON",
+      "relatedSpdxElement": "SPDXRef-Package-pypi-sympy"
+    },
    {
      "spdxElementId": "SPDXRef-Package-one-memory",
      "relationshipType": "DEPENDS_ON",
@@ -2,39 +2,79 @@

 MemHub itself is licensed under Apache-2.0 (see `LICENSE`).
 It bundles the runtime dependencies listed below.
-Generated from the installed dependency tree on 2026-10-15T17:26:47Z
+Generated from the installed dependency tree on 2026-10-17T13:52:48Z
 by `python3 scripts/generate_sbom.py`. Do not edit by hand.

 Per-package license texts are not reproduced here yet; the license
 column is filled in as part of release review, and the SBOM records
 `NOASSERTION` until then rather than guessing.

-## hy-memory 2.0.1
+## hy-memory 2.0.4

-License: MIT
+License: Apache-2.0

 ```text
-MIT License
+Tencent is pleased to support the community by making Hy-Memory available.

-Copyright (c) 2025 HY Memory Contributors
+Copyright (C) 2026 Tencent. All rights reserved.

-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
+Hy-Memory is licensed under Apache-2.0. Hy-Memory does not impose any additional restrictions beyond those specified in the license.

-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
+For the avoidance of doubt, Hy-Memory refers solely to code, parameters, and weights made publicly available by Tencent in accordance with Apache-2.0.

-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+Terms of the Apache-2.0 License:
+--------------------------------------------------------------------
+Apache License
+Version 2.0, January 2004
+http://www.apache.org/licenses/
+
+TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+1. Definitions.
+
+"License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document.
+
+"Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License.
+
+"Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity.
+
+"You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License.
+
+"Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files.
+
+"Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types.
+
+"Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below).
+
+"Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof.
+
+"Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution."
+
+"Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Work.
+
+2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.
+
+3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed.
+
+4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions:
+
+You must give any other recipients of the Work or Derivative Works a copy of this License; and
+You must cause any modified files to carry prominent notices stating that You changed the files; and
+You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and
+If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding those notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License.
+You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License.
+
+5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions.
+
+6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file.
+
+7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License.
+
+8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages.
+
+9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability.
+
+END OF TERMS AND CONDITIONS
 ```

 ## MemTensor memmy-agent (adapted)
@@ -85,7 +125,7 @@ No Memmy brand, icon, font or mascot assets are distributed.
 - tar 7.5.22
 - three 0.185.1

-## Python (PyPI) — 122 packages
+## Python (PyPI) — 126 packages

 - aiohappyeyeballs 2.7.1
 - aiohttp 3.14.3
@@ -104,6 +144,7 @@ No Memmy brand, icon, font or mascot assets are distributed.
 - chromadb 1.5.9
 - click 8.5.0
 - cloudpickle 3.1.2
+- coloredlogs 15.0.1
 - distro 1.9.0
 - durationpy 0.11
 - execnet 2.1.2
@@ -120,6 +161,7 @@ No Memmy brand, icon, font or mascot assets are distributed.
 - httptools 0.8.0
 - httpx 0.28.1
 - huggingface-hub 1.29.0
+- humanfriendly 10.0
 - idna 3.19
 - importlib-resources 7.1.0
 - iniconfig 2.3.0
@@ -142,13 +184,14 @@ No Memmy brand, icon, font or mascot assets are distributed.
 - mdurl 0.1.2
 - mmh3 5.3.0
 - more-itertools 11.1.0
+- mpmath 1.3.0
 - multidict 6.7.1
 - mypy 2.3.1
 - mypy-extensions 1.1.0
 - narwhals 2.25.0
 - numpy 2.5.2
 - oauthlib 3.3.1
- onnxruntime 1.29.0
+- onnxruntime 1.23.2
 - openai 2.54.0
 - opentelemetry-api 1.44.0
 - opentelemetry-exporter-otlp-proto-common 1.44.0
@@ -193,6 +236,7 @@ No Memmy brand, icon, font or mascot assets are distributed.
 - six 1.17.0
 - sniffio 1.3.1
 - starlette 1.6.0
+- sympy 1.14.0
 - tenacity 9.1.4
 - threadpoolctl 3.6.0
 - tiktoken 0.14.0
@@ -1,6 +1,6 @@
 {
  "name": "@one-memory/cli",
-  "version": "1.0.0",
+  "version": "1.0.4",
  "description": "MemHub bootstrap CLI for installing and managing the local-first shared memory runtime",
  "license": "Apache-2.0",
  "repository": {
@@ -13,8 +13,8 @@
 * the real package.json and only strips what is provably not needed at runtime.
 */
 import { execFileSync } from "node:child_process";
+import { mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { createRequire } from "node:module";
-import { readFileSync, writeFileSync, mkdirSync } from "node:fs";
 import { dirname, resolve } from "node:path";
 import { fileURLToPath } from "node:url";

@@ -27,10 +27,7 @@ const releaseVersion = resolveBuildVersion(process.env, repositoryRoot);
 const pkgPath = resolve(root, "package.json");
 const pkg = JSON.parse(readFileSync(pkgPath, "utf8"));
 const publishName = process.env.MEMHUB_NPM_PACKAGE?.trim();
-if (
-  !publishName ||
-  !/^(?:@[a-z0-9._-]+\/)?[a-z0-9._-]+$/.test(publishName)
-) {
+if (!publishName || !/^(?:@[a-z0-9._-]+\/)?[a-z0-9._-]+$/.test(publishName)) {
  throw new Error(
    "MEMHUB_NPM_PACKAGE must name the explicit npm publish identity",
  );
@@ -38,8 +35,12 @@ if (

 const bundle = resolve(root, "dist/one-memory.cjs");
 const source = readFileSync(bundle, "utf8");
-const runtimeRequires = [...source.matchAll(/require\(["']([^"']+)["']\)/g)].map((m) => m[1]);
-const workspaceLeaked = runtimeRequires.filter((name) => name.startsWith("@one-memory/"));
+const runtimeRequires = [
+  ...source.matchAll(/require\(["']([^"']+)["']\)/g),
+].map((m) => m[1]);
+const workspaceLeaked = runtimeRequires.filter((name) =>
+  name.startsWith("@one-memory/"),
+);
 if (workspaceLeaked.length > 0) {
  throw new Error(
    `bundle still requires workspace packages at runtime: ${workspaceLeaked.join(", ")}. ` +
@@ -61,9 +62,7 @@ for (const section of ["dependencies", "devDependencies", "peerDependencies"]) {
      // else is build-time only. The bundle is inlined (tsup noExternal), so
      // in practice only Node builtins remain — but compute it rather than
      // assume, so a future non-inlined dep is not silently dropped.
-      runtimeRequires.some(
-        (req) => req === name || req.startsWith(`${name}/`),
-      ),
+      runtimeRequires.some((req) => req === name || req.startsWith(`${name}/`)),
    ),
  );
  if (Object.keys(published[section]).length === 0) delete published[section];
@@ -77,8 +76,15 @@ published.files = ["dist", "scripts", "package.json"];
 // in the staging dir overrides that and makes the manifest authoritative.
 writeFileSync(
  resolve(outDir, ".npmignore"),
-  ["# Ship the built bundle: the repo root .gitignore excludes dist/.", "!dist/", ""].join("\n"),
+  [
+    "# Ship the built bundle: the repo root .gitignore excludes dist/.",
+    "!dist/",
+    "",
+  ].join("\n"),
 );

-writeFileSync(resolve(outDir, "package.json"), `${JSON.stringify(published, null, 2)}\n`);
+writeFileSync(
+  resolve(outDir, "package.json"),
+  `${JSON.stringify(published, null, 2)}\n`,
+);
 console.log("wrote", resolve(outDir, "package.json"));
@@ -20,11 +20,7 @@ import { pathToFileURL } from "node:url";

 const REQUIRED_ALIASES = {
  daemon: ["start", "stop", "restart", "status"],
-  "startup-service": [
-    "service-plan",
-    "service-install",
-    "service-uninstall",
-  ],
+  "startup-service": ["service-plan", "service-install", "service-uninstall"],
 };

 const REQUIRED_COMMAND_SURFACE = [
@@ -36,7 +32,6 @@ const REQUIRED_COMMAND_SURFACE = [
  "calls",
  "import-file",
  "notes",
-  "permission",
  "extraction",
  "graph",
  "models",
@@ -81,7 +76,7 @@ const REQUIRED_MARKERS = [

 /**
 * Sub-commands the UI-write parity work added. A package can declare the
- * parent command (`permission`, `models`) while still missing the new verb, so
+ * parent command (`models`) while still missing the new verb, so
 * checking command names alone would pass a stale bundle. The marker is the
 * exact literal the catalog usage string contains.
 */
@@ -92,9 +87,6 @@ const REQUIRED_SUBCOMMANDS = [
  { command: "import-file", marker: "memhub import-file discard" },
  { command: "notes", marker: "memhub notes add" },
  { command: "notes", marker: "memhub notes process" },
-  { command: "permission", marker: "memhub permission status" },
-  { command: "permission", marker: "memhub permission allow" },
-  { command: "permission", marker: "memhub permission deny" },
  { command: "extraction", marker: "memhub extraction set" },
  { command: "graph", marker: "memhub graph width" },
  { command: "models", marker: "memhub models edit" },
@@ -112,19 +104,83 @@ const REQUIRED_SUBCOMMANDS = [
 * These run without `--lifecycle`, so they are safe against any package.
 */
 const STOPPED_DESTRUCTIVE_PROBES = [
-  { name: "calls_clear_preview", args: ["calls", "clear", "--event", "11111111-1111-4111-8111-111111111111"] },
-  { name: "calls_clear_apply", args: ["calls", "clear", "--event", "11111111-1111-4111-8111-111111111111", "--confirm"] },
-  { name: "calls_clear_all_apply", args: ["calls", "clear", "--all", "--confirm"] },
-  { name: "import_file_commit_apply", args: ["import-file", "commit", "22222222-2222-4222-8222-222222222222", "--confirm"] },
-  { name: "import_file_discard_apply", args: ["import-file", "discard", "22222222-2222-4222-8222-222222222222", "--confirm"] },
-  { name: "notes_process_apply", args: ["notes", "process", "33333333-3333-4333-8333-333333333333", "--confirm"] },
-  { name: "permission_allow", args: ["permission", "allow"] },
-  { name: "permission_deny", args: ["permission", "deny"] },
-  { name: "extraction_set_apply", args: ["extraction", "set", "--window", "6m", "--confirm"] },
-  { name: "graph_width_apply", args: ["graph", "width", "--decision", "keep_existing", "--confirm"] },
-  { name: "processing_extract_apply", args: ["processing", "extract", "--confirm"] },
-  { name: "models_activate_apply", args: ["models", "activate", "44444444-4444-4444-8444-444444444444", "--confirm"] },
-  { name: "models_edit_apply", args: ["models", "edit", "44444444-4444-4444-8444-444444444444", "--config-file", "/nonexistent/patch.json", "--confirm"] },
+  {
+    name: "calls_clear_preview",
+    args: ["calls", "clear", "--event", "11111111-1111-4111-8111-111111111111"],
+  },
+  {
+    name: "calls_clear_apply",
+    args: [
+      "calls",
+      "clear",
+      "--event",
+      "11111111-1111-4111-8111-111111111111",
+      "--confirm",
+    ],
+  },
+  {
+    name: "calls_clear_all_apply",
+    args: ["calls", "clear", "--all", "--confirm"],
+  },
+  {
+    name: "import_file_commit_apply",
+    args: [
+      "import-file",
+      "commit",
+      "22222222-2222-4222-8222-222222222222",
+      "--confirm",
+    ],
+  },
+  {
+    name: "import_file_discard_apply",
+    args: [
+      "import-file",
+      "discard",
+      "22222222-2222-4222-8222-222222222222",
+      "--confirm",
+    ],
+  },
+  {
+    name: "notes_process_apply",
+    args: [
+      "notes",
+      "process",
+      "33333333-3333-4333-8333-333333333333",
+      "--confirm",
+    ],
+  },
+  {
+    name: "extraction_set_apply",
+    args: ["extraction", "set", "--window", "6m", "--confirm"],
+  },
+  {
+    name: "graph_width_apply",
+    args: ["graph", "width", "--decision", "keep_existing", "--confirm"],
+  },
+  {
+    name: "processing_extract_apply",
+    args: ["processing", "extract", "--confirm"],
+  },
+  {
+    name: "models_activate_apply",
+    args: [
+      "models",
+      "activate",
+      "44444444-4444-4444-8444-444444444444",
+      "--confirm",
+    ],
+  },
+  {
+    name: "models_edit_apply",
+    args: [
+      "models",
+      "edit",
+      "44444444-4444-4444-8444-444444444444",
+      "--config-file",
+      "/nonexistent/patch.json",
+      "--confirm",
+    ],
+  },
 ];

 export function validateCatalog(document) {
@@ -201,7 +257,9 @@ export async function runPackageMatrix(runtimeDist, options = {}) {
  const markers = REQUIRED_MARKERS.filter(
    (marker) => !bundle.toString("utf8").includes(marker),
  );
-  const isolatedRoot = await mkdtemp(path.join(tmpdir(), "memhub-package-matrix-"));
+  const isolatedRoot = await mkdtemp(
+    path.join(tmpdir(), "memhub-package-matrix-"),
+  );
  const home = path.join(isolatedRoot, "home");
  const dataDir = path.join(isolatedRoot, "data");
  const hostHome = path.join(isolatedRoot, "host-home");
@@ -343,7 +401,10 @@ async function runStoppedDestructive(node, cli, environment) {
        `${probe.name} exit ${result.code}, expected 3 while Runtime is stopped`,
      );
    }
-    if (document?.state !== "unknown" || document?.code !== "runtime_not_running") {
+    if (
+      document?.state !== "unknown" ||
+      document?.code !== "runtime_not_running"
+    ) {
      failures.push(
        `${probe.name} did not fail closed with unknown/runtime_not_running`,
      );
@@ -451,8 +512,7 @@ async function main() {
  }
  const asJson = rest.includes("--json");
  const expectShaIndex = rest.indexOf("--expect-sha");
-  const expectSha =
-    expectShaIndex >= 0 ? rest[expectShaIndex + 1] : undefined;
+  const expectSha = expectShaIndex >= 0 ? rest[expectShaIndex + 1] : undefined;
  if (expectSha !== undefined && !/^[0-9a-fA-F]{64}$/.test(expectSha)) {
    process.stderr.write("--expect-sha must be a 64-character SHA-256\n");
    process.exitCode = 2;
@@ -30,12 +30,16 @@ describe("managed restore over HTTP", { timeout: 15_000 }, () => {
      }
      if (request.headers.authorization === "Bearer r409") {
        response.statusCode = 409;
-        response.end(JSON.stringify({ detail: "backup_restore_runtime_running" }));
+        response.end(
+          JSON.stringify({ detail: "backup_restore_runtime_running" }),
+        );
        return;
      }
      if (request.headers.authorization === "Bearer r422") {
        response.statusCode = 422;
-        response.end(JSON.stringify({ detail: "backup_restore_source_outside_home" }));
+        response.end(
+          JSON.stringify({ detail: "backup_restore_source_outside_home" }),
+        );
        return;
      }
      response.end(
@@ -48,9 +52,12 @@ describe("managed restore over HTTP", { timeout: 15_000 }, () => {
        }),
      );
    });
-    await new Promise<void>((resolve) => server.listen(0, "127.0.0.1", resolve));
+    await new Promise<void>((resolve) =>
+      server.listen(0, "127.0.0.1", resolve),
+    );
    const address = server.address();
-    if (address === null || typeof address === "string") throw new Error("no port");
+    if (address === null || typeof address === "string")
+      throw new Error("no port");
    port = address.port;
  });

@@ -137,8 +144,18 @@ describe("maintenance Core identity", { timeout: 15_000 }, () => {

  it("rejects a squatter, wrong service, wrong instance, and non-200", async () => {
    const cases = [
-      { status: 200, data: { status: "ok", service: "other", instance_id: "x" } },
-      { status: 200, data: { status: "ok", service: "one-memory-core", instance_id: "wrong" } },
+      {
+        status: 200,
+        data: { status: "ok", service: "other", instance_id: "x" },
+      },
+      {
+        status: 200,
+        data: {
+          status: "ok",
+          service: "one-memory-core",
+          instance_id: "wrong",
+        },
+      },
      { status: 503, data: {} },
      { status: 200, data: null },
    ];
@@ -91,7 +91,8 @@ export async function callRestore(
  source: string,
  request: typeof defaultPost = defaultPost,
 ): Promise<
-  { state: "restored"; result: RestoreResult } | { state: "rejected"; code: string }
+  | { state: "restored"; result: RestoreResult }
+  | { state: "rejected"; code: string }
 > {
  const response = await request(
    `http://${options.host}:${options.port}/api/v1/backup/restore`,
@@ -1,19 +1,21 @@
 import { describe, expect, it, vi } from "vitest";
-
-import type { CoreReader, ReadOutcome } from "./core-client.js";
-import type { CoreOperations } from "./core-operations.js";
 import {
  batchScanExitCode,
  executeBatchImport,
  planBatchImport,
  renderBatchScanText,
 } from "./batch-import-flow.js";
+import type { CoreReader, ReadOutcome } from "./core-client.js";
+import type { CoreOperations } from "./core-operations.js";

 function readerFor(catalog: unknown): CoreReader {
  return (async (path: string) =>
    path === "/api/v1/sources"
      ? ({ state: "ok", data: catalog } as ReadOutcome<unknown>)
-      : ({ state: "not_found", code: "http_404" } as ReadOutcome<unknown>)) as CoreReader;
+      : ({
+          state: "not_found",
+          code: "http_404",
+        } as ReadOutcome<unknown>)) as CoreReader;
 }

 function found(source_id: string, extra: Record<string, unknown> = {}) {
@@ -81,18 +83,36 @@ describe("A19 batch execution", () => {
      readerFor([found("codex"), found("cursor")]),
    );
    if (planned.state !== "preview") throw new Error("expected preview");
-    const post = vi.fn(async (_path: string, body: Record<string, unknown>) => ({
-      state: "committed" as const,
-      data: {
-        request_id: body.request_id,
-        accepted_source_ids: ["codex", "cursor"],
-        failed_sources: {},
-        jobs: [
-          { job_id: "j1", source_id: "codex", status: "pending", discovered: 3, stored: 3, errors: 0, operation: "history" },
-          { job_id: "j2", source_id: "cursor", status: "pending", discovered: 5, stored: 5, errors: 0, operation: "history" },
-        ],
-      },
-    }));
+    const post = vi.fn(
+      async (_path: string, body: Record<string, unknown>) => ({
+        state: "committed" as const,
+        data: {
+          request_id: body.request_id,
+          accepted_source_ids: ["codex", "cursor"],
+          failed_sources: {},
+          jobs: [
+            {
+              job_id: "j1",
+              source_id: "codex",
+              status: "pending",
+              discovered: 3,
+              stored: 3,
+              errors: 0,
+              operation: "history",
+            },
+            {
+              job_id: "j2",
+              source_id: "cursor",
+              status: "pending",
+              discovered: 5,
+              stored: 5,
+              errors: 0,
+              operation: "history",
+            },
+          ],
+        },
+      }),
+    );
    const operations = { post } as unknown as CoreOperations;
    const document = await executeBatchImport(
      planned.candidates,
@@ -127,7 +147,15 @@ describe("A19 batch execution", () => {
          accepted_source_ids: ["codex"],
          failed_sources: { cursor: "source_identity_changed" },
          jobs: [
-            { job_id: "j1", source_id: "codex", status: "pending", discovered: 1, stored: 1, errors: 0, operation: "history" },
+            {
+              job_id: "j1",
+              source_id: "codex",
+              status: "pending",
+              discovered: 1,
+              stored: 1,
+              errors: 0,
+              operation: "history",
+            },
          ],
        },
      })),
@@ -242,7 +270,15 @@ describe("A19 batch execution", () => {
          accepted_source_ids: ["codex"],
          failed_sources: {},
          jobs: [
-            { job_id: "j1", source_id: "codex", status: "pending", discovered: 1, stored: 1, errors: 0, operation: "history" },
+            {
+              job_id: "j1",
+              source_id: "codex",
+              status: "pending",
+              discovered: 1,
+              stored: 1,
+              errors: 0,
+              operation: "history",
+            },
          ],
        },
      })),
@@ -14,15 +14,14 @@
 */

 import { randomUUID } from "node:crypto";
-import { buildSourceImportRequest } from "@one-memory/operations";
-
 import type { ImportJob } from "@one-memory/contracts";
+import { buildSourceImportRequest } from "@one-memory/operations";
+import type { CoreReader } from "./core-client.js";
+import type { CoreOperations } from "./core-operations.js";
 import {
  type SourceConnectionCandidate,
  validateSourceConnectionCandidates,
 } from "./core-responses.js";
-import type { CoreReader } from "./core-client.js";
-import type { CoreOperations } from "./core-operations.js";
 import { PRODUCT } from "./version.js";

 export interface BatchSourceRow {
@@ -57,19 +56,34 @@ export function detectedSourceIds(
  return candidates.map((candidate) => candidate.source_id);
 }

-export async function planBatchImport(
-  reader: CoreReader,
-): Promise<
-  | { state: "preview"; candidates: SourceConnectionCandidate[]; rows: BatchSourceRow[]; notFound: string[] }
-  | { state: "rejected" | "unknown" | "no_sources"; code: string; notFound: string[] }
+export async function planBatchImport(reader: CoreReader): Promise<
+  | {
+      state: "preview";
+      candidates: SourceConnectionCandidate[];
+      rows: BatchSourceRow[];
+      notFound: string[];
+    }
+  | {
+      state: "rejected" | "unknown" | "no_sources";
+      code: string;
+      notFound: string[];
+    }
 > {
  const outcome = await reader<unknown>("/api/v1/sources");
  if (outcome.state !== "ok") {
-    return { state: outcome.state === "unavailable" ? "unknown" : "rejected", code: outcome.code, notFound: [] };
+    return {
+      state: outcome.state === "unavailable" ? "unknown" : "rejected",
+      code: outcome.code,
+      notFound: [],
+    };
  }
  const catalog = outcome.data;
  if (!Array.isArray(catalog) || catalog.length > SOURCE_LIST_LIMIT) {
-    return { state: "rejected", code: "sources_response_invalid", notFound: [] };
+    return {
+      state: "rejected",
+      code: "sources_response_invalid",
+      notFound: [],
+    };
  }
  const candidates = validateSourceConnectionCandidates(catalog);
  if (!candidates) {
@@ -165,9 +179,7 @@ export async function executeBatchImport(
    receipt.failed_sources && typeof receipt.failed_sources === "object"
      ? (receipt.failed_sources as Record<string, string>)
      : {};
-  const jobs = Array.isArray(receipt.jobs)
-    ? (receipt.jobs as ImportJob[])
-    : [];
+  const jobs = Array.isArray(receipt.jobs) ? (receipt.jobs as ImportJob[]) : [];
  const firstSeen = jobs.map((job) => ({
    source_id: job.source_id,
    discovered: job.discovered,
@@ -262,7 +274,8 @@ export async function executeBatchImport(
 }

 export function batchScanExitCode(document: BatchScanDocument): number {
-  if (document.state === "committed" || document.state === "no_sources") return 0;
+  if (document.state === "committed" || document.state === "no_sources")
+    return 0;
  if (document.state === "partial") return 1;
  return 1;
 }
@@ -81,6 +81,7 @@ import {
  readMemoryWrite,
  readScanRoots,
  removeScanRoot,
+  resetOnboarding,
  submitBackup,
  submitMemoryWrite,
  submitScan,
@@ -190,13 +191,6 @@ import {
  renderMemoryDetailText,
  renderMemoryPageText,
 } from "./memory-report.js";
-import {
-  type ModelContentPermissionDocument,
-  modelContentPermissionExitCode,
-  readModelContentPermission,
-  renderModelContentPermissionText,
-  setModelContentPermission,
-} from "./model-content-permission-flow.js";
 import {
  executeModelProbe,
  inspectModelProbePlan,
@@ -217,12 +211,10 @@ import {
  executeModelProfileActivate,
  executeModelProfileEdit,
  type ModelWriteDocument,
-  type ModelWritePlan,
  modelWriteExitCode,
  planModelProfileActivate,
  planModelProfileEdit,
  renderModelWriteText,
-  validateModelWritePlan,
 } from "./model-write-flows.js";
 import { claimPlanAttempt } from "./operation-journal.js";
 import {
@@ -697,6 +689,36 @@ async function runCommand(
  // `path list` and `adapter list` are read-only: they have no plan to confirm,
  // so they are answered here rather than falling into the two-phase branch
  // (which would reject them for having nothing to apply).
+  // Back to the cold-start screens on purpose. Deleting the app cannot do this
+  // (the completion flag lives with the memories, by design), and deleting the
+  // data to get there would cost the user everything they have stored.
+  if (named === "onboarding" && arguments_.includes("reset")) {
+    if (
+      arguments_.filter((item) => item === "--json").length > 1 ||
+      arguments_.filter((item) => item !== "--json").join(" ") !==
+        "onboarding reset"
+    ) {
+      return usage();
+    }
+    const attachment = await attachToRuntime(
+      resolveRuntimeCliConfig(environment, process.argv[1]).dataDir,
+    );
+    if (attachment === null) {
+      return runtimeUnavailable(descriptor, asJson);
+    }
+    const state = await resetOnboarding(attachment);
+    if (state === null) {
+      return reject(descriptor, asJson, "引导状态端点不可用。");
+    }
+    if (asJson) {
+      process.stdout.write(`${JSON.stringify(state)}\n`);
+      return EXIT_OK;
+    }
+    process.stdout.write(
+      `引导流程已回到首次运行（status=${state.status}），记忆数据未改动。\n`,
+    );
+    return EXIT_OK;
+  }
  if (named === "path" && arguments_.includes("list")) {
    if (
      arguments_.filter((item) => item === "--json").length > 1 ||
@@ -2462,45 +2484,6 @@ async function runCommand(
    );
    return uiWriteExitCode(document);
  }
-  if (named === "permission") {
-    const parsed = parsePermissionCommand(arguments_.slice(1));
-    if (!parsed) return usage();
-    const config = resolveRuntimeCliConfig(environment, process.argv[1]);
-    const runtime = await inspectRuntimeStatus({
-      dataDir: config.dataDir,
-      probeIdentity,
-      requireServiceReceipt: true,
-    });
-    const attached =
-      runtime.state === "running"
-        ? await attachClients(config.dataDir)
-        : { reader: null, operations: null, failure: runtime.code };
-    if (
-      runtime.state !== "running" ||
-      (parsed === "status" ? !attached.reader : !attached.operations)
-    ) {
-      const document: ModelContentPermissionDocument = {
-        schema_version: 1,
-        product: PRODUCT,
-        state: "unknown",
-        code:
-          runtime.state !== "running"
-            ? runtime.code
-            : (attached.failure ?? "runtime_attach_failed"),
-      };
-      emitModelContentPermissionDocument(document, asJson);
-      return runtime.state === "stopped" ? 3 : 1;
-    }
-    const document =
-      parsed === "status"
-        ? await readModelContentPermission(attached.reader!)
-        : await setModelContentPermission(
-            attached.operations!,
-            parsed === "allow",
-          );
-    emitModelContentPermissionDocument(document, asJson);
-    return modelContentPermissionExitCode(document);
-  }
  if (named === "notes" || named === "extraction" || named === "graph") {
    const parsed = parseExtendedWriteCommand(
      named as "notes" | "extraction" | "graph",
@@ -5124,18 +5107,6 @@ function parseHeadlessCommand(
    : null;
 }

-function parsePermissionCommand(
-  rest: readonly string[],
-): "status" | "allow" | "deny" | null {
-  if (rest.filter((item) => item === "--json").length > 1) return null;
-  const tokens = rest.filter((item) => item !== "--json");
-  const action = tokens[0];
-  return tokens.length === 1 &&
-    (action === "status" || action === "allow" || action === "deny")
-    ? action
-    : null;
-}
-
 function unwrapPlan(value: unknown): unknown {
  if (
    value &&
@@ -5323,17 +5294,6 @@ function emitModelReportDocument(
  );
 }

-function emitModelContentPermissionDocument(
-  document: ModelContentPermissionDocument,
-  asJson: boolean,
-): void {
-  process.stdout.write(
-    asJson
-      ? `${JSON.stringify(document)}\n`
-      : renderModelContentPermissionText(document),
-  );
-}
-
 function emitSourceSyncDocument(
  document: SourceSyncFlowDocument,
  asJson: boolean,
@@ -282,11 +282,7 @@ export const COMMANDS: readonly CommandDescriptor[] = [
    summary: "列出来源、执行已确认的导入计划。",
    usage:
      "memhub sources list [--json]\n            memhub sources import --plan-file <path> --confirm [--json]\n            memhub sources receipt <request-id> [--json]",
-    options: [
-      PLAN_FILE_OPTION,
-      EXPLICIT_CONFIRM_OPTION,
-      JSON_OPTION,
-    ],
+    options: [PLAN_FILE_OPTION, EXPLICIT_CONFIRM_OPTION, JSON_OPTION],
    exit_codes: [EXIT_OK, EXIT_FAILED, EXIT_USAGE, EXIT_RUNTIME_STOPPED],
  },
  {
@@ -497,14 +493,6 @@ export const COMMANDS: readonly CommandDescriptor[] = [
    ],
    exit_codes: [EXIT_OK, EXIT_FAILED, EXIT_USAGE, EXIT_RUNTIME_STOPPED],
  },
-  {
-    name: "permission",
-    summary: "读取或直接设置全局模型内容外发许可。",
-    usage:
-      "memhub permission status [--json]\n            memhub permission allow [--json]\n            memhub permission deny [--json]",
-    options: [JSON_OPTION],
-    exit_codes: [EXIT_OK, EXIT_FAILED, EXIT_USAGE, EXIT_RUNTIME_STOPPED],
-  },
  {
    name: "extraction",
    summary: "设置提取窗口；默认只出计划，--confirm 后才写入。",
@@ -663,6 +651,14 @@ export const COMMANDS: readonly CommandDescriptor[] = [
    ],
    exit_codes: [EXIT_OK, EXIT_FAILED, EXIT_USAGE, EXIT_PLAN_BLOCKED],
  },
+  {
+    name: "onboarding",
+    summary:
+      "onboarding reset 回到首次运行的引导流程。只清除引导完成状态：不删记忆数据，也不需要卸载重装。",
+    usage: "memhub onboarding reset [--json]",
+    options: [JSON_OPTION],
+    exit_codes: [EXIT_OK, EXIT_FAILED, EXIT_USAGE],
+  },
  {
    name: "help",
    summary: "显示总帮助或单个命令的用法。",
@@ -81,7 +81,6 @@ const COMMANDS = [
  "calls",
  "import-file",
  "notes",
-  "permission",
  "extraction",
  "graph",
  "host",
@@ -93,6 +92,7 @@ const COMMANDS = [
  "uninstall",
  "upgrade",
  "delete-data",
+  "onboarding",
  "help",
  "version",
  // A6 product commands. Appended, not inserted: the catalog is a stable
@@ -259,7 +259,14 @@ describe("CLI-22 shared output contract", () => {

  describe("argument validation", () => {
    it("fails unknown commands without printing data or running anything", async () => {
-      for (const argv of [["unknown"], ["unknown", "--help"], ["--typo"]]) {
+      for (const argv of [
+        ["unknown"],
+        ["unknown", "--help"],
+        ["--typo"],
+        ["permission", "status"],
+        ["permission", "allow"],
+        ["permission", "deny"],
+      ]) {
        const streams = captureStreams();
        try {
          expect(await run(argv, {})).toBe(2);
@@ -300,11 +307,7 @@ describe("CLI-22 shared output contract", () => {
          `${PRODUCT} sync`,
        ],
        [
-          [
-            "sync",
-            "receipt",
-            "123e4567-e89b-12d3-a456-426614174000",
-          ],
+          ["sync", "receipt", "123e4567-e89b-12d3-a456-426614174000"],
          `${PRODUCT} sync`,
        ],
        [["sync", "pause", "--source"], `${PRODUCT} sync`],
@@ -555,12 +558,7 @@ describe("documented commands", () => {
    // Source processing is one Freeze/Thaw switch under `sync`; the obsolete
    // independent read/write `sources access` command must not return here.
    sources: ["list", "import", "receipt"],
-    jobs: [
-      "list",
-      "show",
-      "start",
-      "<pause|resume|cancel>",
-    ],
+    jobs: ["list", "show", "start", "<pause|resume|cancel>"],
    // `sync <pause|resume>` is documented as one alternation, so it is
    // checked as an entry rather than as two literal subcommands.
    sync: ["status", "<pause|resume>"],
@@ -342,17 +342,20 @@ describe("unified context search", () => {
    ["--limit", "1", "2"],
    ["--character-budget", "200", "300"],
    ["--timeout-ms", "100", "200"],
-  ])("rejects a repeated %s before Runtime attach", async (flag, first, second) => {
-    const streams = captureStreams();
-    try {
-      expect(
-        await run(["recall", "q", flag, first, flag, second, "--json"], {}),
-      ).toBe(2);
-    } finally {
-      streams.restore();
-    }
-    expect(streams.stdout()).toBe("");
-  });
+  ])(
+    "rejects a repeated %s before Runtime attach",
+    async (flag, first, second) => {
+      const streams = captureStreams();
+      try {
+        expect(
+          await run(["recall", "q", flag, first, flag, second, "--json"], {}),
+        ).toBe(2);
+      } finally {
+        streams.restore();
+      }
+      expect(streams.stdout()).toBe("");
+    },
+  );

  it("refuses malformed success bodies instead of leaking nested values", async () => {
    transport.post.mockResolvedValue({
@@ -381,6 +384,101 @@ describe("unified context search", () => {
    expect(streams.stdout()).not.toContain("must-not-print");
  });

+  it("counts characters as code points so an emoji hit stays grounded", async () => {
+    // Core reports used_characters with Python's len(), which counts code
+    // points: this text is 10 characters there and 12 UTF-16 units here.
+    const text = "消费保 📍 扩展 📡";
+    expect(text.length).toBe(12);
+    expect(Array.from(text).length).toBe(10);
+    transport.post.mockResolvedValue({
+      status: 200,
+      data: {
+        ...RESPONSE,
+        paths: {
+          ...RESPONSE.paths,
+          memory: {
+            ...RESPONSE.paths.memory,
+            status: "empty",
+            returned_count: 0,
+          },
+        },
+        items: [
+          {
+            ...(RESPONSE.items[0] as Record<string, unknown>),
+            messages: [
+              {
+                ...(
+                  RESPONSE.items[0] as {
+                    messages: Array<Record<string, unknown>>;
+                  }
+                ).messages[0],
+                text,
+              },
+            ],
+          },
+        ],
+        used_characters: 10,
+      },
+    });
+    const streams = captureStreams();
+    try {
+      expect(await run(["recall", "test", "--json"])).toBe(0);
+    } finally {
+      streams.restore();
+    }
+    expect(JSON.parse(streams.stdout())).toMatchObject({
+      state: "grounded",
+      used_characters: 10,
+      items: [{ kind: "raw", messages: [{ text }] }],
+    });
+  });
+
+  it("still rejects a response whose character total does not match", async () => {
+    // The same text counted as UTF-16 units is exactly what Core does not mean,
+    // so the cross-check must keep failing instead of accepting both totals.
+    const text = "消费保 📍 扩展 📡";
+    transport.post.mockResolvedValue({
+      status: 200,
+      data: {
+        ...RESPONSE,
+        paths: {
+          ...RESPONSE.paths,
+          memory: {
+            ...RESPONSE.paths.memory,
+            status: "empty",
+            returned_count: 0,
+          },
+        },
+        items: [
+          {
+            ...(RESPONSE.items[0] as Record<string, unknown>),
+            messages: [
+              {
+                ...(
+                  RESPONSE.items[0] as {
+                    messages: Array<Record<string, unknown>>;
+                  }
+                ).messages[0],
+                text,
+              },
+            ],
+          },
+        ],
+        used_characters: 12,
+      },
+    });
+    const streams = captureStreams();
+    try {
+      expect(await run(["recall", "test", "--json"])).toBe(1);
+    } finally {
+      streams.restore();
+    }
+    expect(JSON.parse(streams.stdout())).toMatchObject({
+      state: "invalid",
+      code: "response_invalid",
+    });
+  });
+
  it("reports endpoint denial without exposing the response body", async () => {
    transport.post.mockResolvedValue({
      status: 403,
@@ -181,9 +181,9 @@ function validateContextResponse(
    items.push(item);
    characters +=
      item.kind === "memory"
-        ? String(item.content).length
+        ? characterCount(item.content)
        : (item.messages as Array<Record<string, unknown>>).reduce(
-            (sum, message) => sum + String(message.text).length,
+            (sum, message) => sum + characterCount(message.text),
            0,
          );
  }
@@ -330,6 +330,17 @@ function record(value: unknown): Record<string, unknown> | null {
    : null;
 }

+/**
+ * Count characters the way Core does. Python's len() counts Unicode code
+ * points; String.prototype.length counts UTF-16 code units, so an emoji outside
+ * the BMP is one character to Core and two here. Comparing the two totals made
+ * every response containing an emoji fail the used_characters cross-check and
+ * surface as response_invalid.
+ */
+function characterCount(value: unknown): number {
+  return Array.from(String(value)).length;
+}
+
 function boundedText(
  value: unknown,
  maximum: number,
@@ -1,5 +1,5 @@
-import axios from "axios";
 import { classifyModelRemove } from "@one-memory/operations";
+import axios from "axios";

 /**
 * Only paths proven side-effect free by reading the Core handlers belong here.
@@ -8,7 +8,6 @@ import { classifyModelRemove } from "@one-memory/operations";
 */
 const READ_ONLY_PATHS = [
  "/api/v1/call-events",
-  "/api/v1/model-content-permission",
  "/api/v1/home-summary",
  "/api/v1/import-jobs",
  "/api/v1/live-sync",
@@ -56,7 +55,10 @@ export function createUiSessionRequest(attachment: {
      return { state: "ok" as const, data: response.data };
    }
    if (response.status === 503) {
-      return { state: "unavailable" as const, code: "ui_web_assets_unavailable" };
+      return {
+        state: "unavailable" as const,
+        code: "ui_web_assets_unavailable",
+      };
    }
    return { state: "unavailable" as const, code: `http_${response.status}` };
  };
@@ -395,15 +397,30 @@ export async function readMemoryWrite(
  }
 }

-export async function readScanRoots(
+export interface OnboardingState {
+  status: string;
+  selected_source_ids: string[];
+  started_at: string | null;
+  completed_at: string | null;
+}
+
+/** Send onboarding back to its first-run state.
+ *
+ * Returns null when the endpoint is not there, so an older runtime degrades to
+ * "not available" instead of failing the command. Only the completion flag
+ * moves: memories, sources and settings live in other tables, which is what
+ * makes this the honest way to see the cold-start screens again.
+ */
+export async function resetOnboarding(
  attachment: RuntimeAttachmentLike,
-): Promise<{ schema_version: number; revision: number; roots: unknown[] } | null> {
+): Promise<OnboardingState | null> {
  try {
-    const response = await axios.get<{ schema_version: number; revision: number; roots: unknown[] }>(
-      `${baseUrlOf(attachment)}/api/v1/scan-roots`,
+    const response = await axios.post<OnboardingState>(
+      `${baseUrlOf(attachment)}/api/v1/onboarding/reset`,
+      undefined,
      {
        headers: { Authorization: `Bearer ${attachment.apiToken}` },
-        timeout: 10_000,
+        timeout: 15_000,
        proxy: false,
        maxRedirects: 0,
        maxContentLength: MAX_RESPONSE_BYTES,
@@ -415,12 +432,19 @@ export async function readScanRoots(
  }
 }

-export async function addScanRoot(
+export async function readScanRoots(
  attachment: RuntimeAttachmentLike,
-  body: { root: string; source?: string; recursive?: boolean },
-): Promise<unknown | null> {
+): Promise<{
+  schema_version: number;
+  revision: number;
+  roots: unknown[];
+} | null> {
  try {
-    const response = await axios.post<unknown>(`${baseUrlOf(attachment)}/api/v1/scan-roots`, body, {
+    const response = await axios.get<{
+      schema_version: number;
+      revision: number;
+      roots: unknown[];
+    }>(`${baseUrlOf(attachment)}/api/v1/scan-roots`, {
      headers: { Authorization: `Bearer ${attachment.apiToken}` },
      timeout: 10_000,
      proxy: false,
@@ -428,6 +452,28 @@ export async function addScanRoot(
      maxContentLength: MAX_RESPONSE_BYTES,
    });
    return response.data;
+  } catch {
+    return null;
+  }
+}
+
+export async function addScanRoot(
+  attachment: RuntimeAttachmentLike,
+  body: { root: string; source?: string; recursive?: boolean },
+): Promise<unknown | null> {
+  try {
+    const response = await axios.post<unknown>(
+      `${baseUrlOf(attachment)}/api/v1/scan-roots`,
+      body,
+      {
+        headers: { Authorization: `Bearer ${attachment.apiToken}` },
+        timeout: 10_000,
+        proxy: false,
+        maxRedirects: 0,
+        maxContentLength: MAX_RESPONSE_BYTES,
+      },
+    );
+    return response.data;
  } catch (error) {
    return { _error: error instanceof Error ? error.message : String(error) };
  }
@@ -438,14 +484,17 @@ export async function removeScanRoot(
  body: { root: string },
 ): Promise<unknown | null> {
  try {
-    const response = await axios.delete<unknown>(`${baseUrlOf(attachment)}/api/v1/scan-roots`, {
-      headers: { Authorization: `Bearer ${attachment.apiToken}` },
-      data: body,
-      timeout: 10_000,
-      proxy: false,
-      maxRedirects: 0,
-      maxContentLength: MAX_RESPONSE_BYTES,
-    });
+    const response = await axios.delete<unknown>(
+      `${baseUrlOf(attachment)}/api/v1/scan-roots`,
+      {
+        headers: { Authorization: `Bearer ${attachment.apiToken}` },
+        data: body,
+        timeout: 10_000,
+        proxy: false,
+        maxRedirects: 0,
+        maxContentLength: MAX_RESPONSE_BYTES,
+      },
+    );
    return response.data;
  } catch (error) {
    return { _error: error instanceof Error ? error.message : String(error) };
@@ -456,16 +505,16 @@ export async function readAdapters(
  attachment: RuntimeAttachmentLike,
 ): Promise<{ schema_version: number; adapters: unknown[] } | null> {
  try {
-    const response = await axios.get<{ schema_version: number; adapters: unknown[] }>(
-      `${baseUrlOf(attachment)}/api/v1/adapters`,
-      {
-        headers: { Authorization: `Bearer ${attachment.apiToken}` },
-        timeout: 10_000,
-        proxy: false,
-        maxRedirects: 0,
-        maxContentLength: MAX_RESPONSE_BYTES,
-      },
-    );
+    const response = await axios.get<{
+      schema_version: number;
+      adapters: unknown[];
+    }>(`${baseUrlOf(attachment)}/api/v1/adapters`, {
+      headers: { Authorization: `Bearer ${attachment.apiToken}` },
+      timeout: 10_000,
+      proxy: false,
+      maxRedirects: 0,
+      maxContentLength: MAX_RESPONSE_BYTES,
+    });
    return response.data;
  } catch {
    return null;
@@ -30,7 +30,6 @@ const IMPORT_FILE =
 const MEMORY_NOTES_PATH = "/api/v1/memory-notes";
 const MEMORY_NOTE_PROCESS =
  /^\/api\/v1\/memory-notes\/[A-Za-z0-9][\w.:-]{0,127}\/process$/;
-const MODEL_CONTENT_PERMISSION_PATH = "/api/v1/model-content-permission";
 const EXTRACTION_SETTINGS_PATH = "/api/v1/extraction-settings";
 const GRAPH_WIDTH_PATH = "/api/v1/graph-width";
 const MANUAL_EXTRACT_PATH = "/api/v1/memory-processing/manual-extract";
@@ -134,7 +133,6 @@ export function createCoreOperations(attachment: {
        !IMPORT_FILE_COMMIT.test(path) &&
        path !== MEMORY_NOTES_PATH &&
        !MEMORY_NOTE_PROCESS.test(path) &&
-        path !== MODEL_CONTENT_PERMISSION_PATH &&
        path !== GRAPH_WIDTH_PATH &&
        path !== MANUAL_EXTRACT_PATH &&
        path !== ONBOARDING_COMPLETE_PATH &&
@@ -1,4 +1,4 @@
-import { mkdtemp, mkdir, readFile, stat, writeFile } from "node:fs/promises";
+import { mkdir, mkdtemp, readFile, stat, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
 import { describe, expect, it } from "vitest";
@@ -23,14 +23,22 @@ describe("explicit data deletion", () => {
      "one-memory.sqlite3": "db",
      "runtime-service.json": "{}",
    });
-    const plan = await planDataDeletion({ dataDir, home, runtimeRunning: false });
+    const plan = await planDataDeletion({
+      dataDir,
+      home,
+      runtimeRunning: false,
+    });
    expect(plan.state).toBe("preview");
    expect(plan.entries).toContain(path.join(dataDir, "one-memory.sqlite3"));
  });

  it("refuses while the Runtime is still running", async () => {
    const { home, dataDir } = await homeWith();
-    const plan = await planDataDeletion({ dataDir, home, runtimeRunning: true });
+    const plan = await planDataDeletion({
+      dataDir,
+      home,
+      runtimeRunning: true,
+    });
    expect(plan).toMatchObject({
      state: "blocked",
      code: "runtime_still_running",
@@ -66,16 +74,23 @@ describe("explicit data deletion", () => {
        home,
        runtimeRunning: false,
      });
-      expect(plan).toMatchObject({ state: "blocked", code: "data_dir_refused" });
+      expect(plan).toMatchObject({
+        state: "blocked",
+        code: "data_dir_refused",
+      });
    }
  });

  it("requires the second confirmation to name the exact directory", async () => {
    const { home, dataDir } = await homeWith();
-    const plan = await planDataDeletion({ dataDir, home, runtimeRunning: false });
-    await expect(applyDataDeletion(plan, path.join(home, "other"))).rejects.toThrow(
-      "data_deletion_confirmation_mismatch",
-    );
+    const plan = await planDataDeletion({
+      dataDir,
+      home,
+      runtimeRunning: false,
+    });
+    await expect(
+      applyDataDeletion(plan, path.join(home, "other")),
+    ).rejects.toThrow("data_deletion_confirmation_mismatch");
    // The data survived a mismatched confirmation.
    await expect(
      stat(path.join(dataDir, "one-memory.sqlite3")),
@@ -84,7 +99,11 @@ describe("explicit data deletion", () => {

  it("deletes an owned Home only when both confirmations match", async () => {
    const { home, dataDir } = await homeWith({ "one-memory.sqlite3": "db" });
-    const plan = await planDataDeletion({ dataDir, home, runtimeRunning: false });
+    const plan = await planDataDeletion({
+      dataDir,
+      home,
+      runtimeRunning: false,
+    });
    const receipt = await applyDataDeletion(plan, dataDir);
    expect(receipt.state).toBe("deleted");
    await expect(stat(dataDir)).rejects.toMatchObject({ code: "ENOENT" });
@@ -68,7 +68,8 @@ export function renderDoctorText(document: DoctorDocument): string {
  return [
    `MemHub 检查：${doctorState(document.state)}`,
    ...document.checks.map(
-      (check) => `${symbol(check.state)} ${labels[check.id]}：${checkState(check.state)}`,
+      (check) =>
+        `${symbol(check.state)} ${labels[check.id]}：${checkState(check.state)}`,
    ),
    ...(document.state === "healthy"
      ? []
@@ -28,11 +28,7 @@ export function resolveEntryDiagnostics(input: {
  // comparison would misread that as isolated and silently drop diagnostics.
  const home = resolve(input.home);
  const accountHome = resolve(input.accountHome);
-  if (
-    home === undefined ||
-    accountHome === undefined ||
-    home === accountHome
-  ) {
+  if (home === undefined || accountHome === undefined || home === accountHome) {
    return { kind: "log" };
  }
  // Isolated HOME with no explicit data dir: the fallback path would write
@@ -9,8 +9,8 @@ import {
  resolveRuntimeCliConfig,
 } from "@one-memory/runtime-control";
 import { run } from "./cli.js";
-import { resolveDiagnosticsDataDir } from "./install-receipt.js";
 import { resolveEntryDiagnostics } from "./entry-diagnostics.js";
+import { resolveDiagnosticsDataDir } from "./install-receipt.js";

 function accountHomeDirectory(): string | undefined {
  try {
@@ -1,12 +1,12 @@
 import { createHash, randomUUID } from "node:crypto";
+import type { ImportJob, ModelProfile } from "@one-memory/contracts";
 import {
-  ModelProfileInputError,
-  SourceImportInputError,
  buildModelProfileCreateRequest,
  buildSourceImportRequest,
+  ModelProfileInputError,
  parseModelProfileReceipt,
+  SourceImportInputError,
 } from "@one-memory/operations";
-import type { ImportJob, ModelProfile } from "@one-memory/contracts";
 import type { CoreReader, ReadOutcome } from "./core-client.js";
 import type { CoreOperations } from "./core-operations.js";
 import {
@@ -117,9 +117,7 @@ function classifySourceImport(
      product: PRODUCT,
      state: "rejected",
      operation_id: operationId,
-      code: onlyBusy
-        ? "source_import_all_busy"
-        : "source_import_all_rejected",
+      code: onlyBusy ? "source_import_all_busy" : "source_import_all_rejected",
      receipt,
    };
  }
@@ -266,7 +266,9 @@ export function renderHostText(document: HostFlowDocument): string {
      }
      if (review.length > 12) lines.push(`    … 另有 ${review.length - 12} 处`);
      lines.push("");
-      lines.push("这些不是我们安装的内容，卸载不会替你删除；需自行处理后才能卸载。");
+      lines.push(
+        "这些不是我们安装的内容，卸载不会替你删除；需自行处理后才能卸载。",
+      );
    }
  }
  return `${lines.join("\n")}\n`;
@@ -259,7 +259,7 @@ describe("A2 dual-mode init", () => {
    expect(streams.value()).toContain("处理：已开启");
  });

-  it("continues the TTY wizard through model probe and content permission", async () => {
+  it("continues the TTY wizard through model setup and probe", async () => {
    const profileId = "22222222-2222-4222-8222-222222222222";
    const profile = {
      profile_id: profileId,
@@ -361,12 +361,6 @@ describe("A2 dual-mode init", () => {
          };
        }
        if (url.endsWith("/activate")) return { status: 200, data: profile };
-        if (url === "/api/v1/model-content-permission") {
-          return {
-            status: 200,
-            data: { allowed: body.allowed },
-          };
-        }
        if (url === "/api/v1/onboarding/complete") {
          return {
            status: 200,
@@ -383,7 +377,7 @@ describe("A2 dual-mode init", () => {
      },
    );
    const prompt = extendedPrompter(
-      ["", "", "", "", "", "", "", "", "n"],
+      ["", "", "", "", "", "", "", "n"],
      "sk-never-print-this",
    );
    const streams = captureStdout();
@@ -402,7 +396,6 @@ describe("A2 dual-mode init", () => {
      streams.restore();
    }
    expect(streams.value()).toContain("已接入来源：codex");
-    expect(prompt.written.join("\n")).toContain("模型内容外发：已允许");
    expect(streams.value()).toContain("首次设置完成。");
    expect(`${streams.value()}\n${prompt.written.join("\n")}`).not.toContain(
      "sk-never-print-this",
@@ -412,10 +405,10 @@ describe("A2 dual-mode init", () => {
      expect.objectContaining({ api_key: "sk-never-print-this" }),
      undefined,
    );
-    expect(transport.post).toHaveBeenCalledWith(
+    expect(transport.post).not.toHaveBeenCalledWith(
      "/api/v1/model-content-permission",
-      { allowed: true },
-      undefined,
+      expect.anything(),
+      expect.anything(),
    );
    expect(transport.post).toHaveBeenCalledWith(
      "/api/v1/onboarding/complete",
@@ -7,7 +7,9 @@ import {
  type WizardPrompter,
 } from "./init-wizard.js";

-function scripted(lines: string[]): WizardPrompter & { asked: string[]; wrote: string[] } {
+function scripted(
+  lines: string[],
+): WizardPrompter & { asked: string[]; wrote: string[] } {
  let index = 0;
  const asked: string[] = [];
  const wrote: string[] = [];
@@ -99,8 +99,7 @@ export async function runInitWizard(
    "开启处理？开启后会处理历史未完成会话并持续接收新会话；输入 n 保持冻结。[Y/n]: ",
  );
  const processing = parseYesNo(processingAnswer, true);
-  if (processing === null)
-    return { state: "invalid", code: "invalid_answer" };
+  if (processing === null) return { state: "invalid", code: "invalid_answer" };

  return {
    state: "answered",
@@ -1,15 +1,15 @@
 import {
  mkdir,
  mkdtemp,
-  readFile,
  readdir,
+  readFile,
  rm,
  writeFile,
 } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
 import { describe, expect, it } from "vitest";
-
+import { resolveEntryDiagnostics } from "./entry-diagnostics.js";
 import {
  defaultInstallReceiptPath,
  readInstallReceipt,
@@ -18,7 +18,6 @@ import {
  resolveInstalledDataDir,
  writeInstallReceipt,
 } from "./install-receipt.js";
-import { resolveEntryDiagnostics } from "./entry-diagnostics.js";

 async function root(): Promise<string> {
  return mkdtemp(path.join(tmpdir(), "memhub-install-receipt-"));
@@ -230,9 +229,9 @@ describe("managed install receipt", () => {
      }),
    ).toBe("/tmp/env data");
    // Receipt beats default.
-    expect(
-      await resolveDiagnosticsDataDir(["install"], { HOME: home }),
-    ).toBe(path.join(home, "receipt data"));
+    expect(await resolveDiagnosticsDataDir(["install"], { HOME: home })).toBe(
+      path.join(home, "receipt data"),
+    );
    // Invalid explicit flag must not silently fall back to the receipt.
    expect(
      await resolveDiagnosticsDataDir(
@@ -132,12 +132,6 @@ function clients(options: { failedProbes?: number } = {}) {
      if (path.endsWith("/activate")) {
        return { state: "committed" as const, data: profile(true) };
      }
-      if (path === "/api/v1/model-content-permission") {
-        return {
-          state: "committed" as const,
-          data: { allowed: body.allowed },
-        };
-      }
      return {
        state: "rejected" as const,
        code: "unexpected_path",
@@ -160,8 +154,8 @@ function clients(options: { failedProbes?: number } = {}) {
 }

 describe("interactive model setup", () => {
-  it("uses App defaults, keeps the key out of output, and allows model content", async () => {
-    const prompt = prompter(["", "", "", "", "", ""]);
+  it("uses App defaults and keeps the key out of output", async () => {
+    const prompt = prompter(["", "", ""]);
    const api = clients();

    const result = await runInteractiveModelSetup(prompt.value, api, {
@@ -170,10 +164,7 @@ describe("interactive model setup", () => {
      installHosts: false,
    });

-    expect(result).toMatchObject({
-      state: "configured",
-      model_content_allowed: true,
-    });
+    expect(result).toMatchObject({ state: "configured" });
    expect(api.post).toHaveBeenCalledWith(
      "/api/v1/model-profiles",
      expect.objectContaining({
@@ -184,9 +175,9 @@ describe("interactive model setup", () => {
        embedding_provider_kind: "local_onnx",
      }),
    );
-    expect(api.post).toHaveBeenCalledWith(
+    expect(api.post).not.toHaveBeenCalledWith(
      "/api/v1/model-content-permission",
-      { allowed: true },
+      expect.anything(),
    );
    expect(prompt.written.join("\n")).not.toContain("sk-do-not-print");
  });
@@ -206,7 +197,7 @@ describe("interactive model setup", () => {
  });

  it("removes an unverified profile before skipping after a failed probe", async () => {
-    const prompt = prompter(["", "", "", "n", "0"]);
+    const prompt = prompter(["", "", "", "0"]);
    const api = clients();
    api.post.mockImplementation(
      async (path: string, body: Record<string, unknown>) => {
@@ -281,7 +272,7 @@ describe("interactive model setup", () => {
  });

  it("cleans a failed probe and lets the user re-enter the model settings", async () => {
-    const prompt = prompter(["", "", "", "", "", "", "", "", "", "", ""]);
+    const prompt = prompter(["", "", "", "", "", "", ""]);
    const api = clients({ failedProbes: 1 });

    const result = await runInteractiveModelSetup(prompt.value, api, {
@@ -290,10 +281,7 @@ describe("interactive model setup", () => {
      installHosts: false,
    });

-    expect(result).toMatchObject({
-      state: "configured",
-      model_content_allowed: true,
-    });
+    expect(result).toMatchObject({ state: "configured" });
    expect(api.remove).toHaveBeenCalledTimes(1);
    expect(
      api.post.mock.calls.filter(([path]) => path === "/api/v1/model-profiles"),
@@ -306,7 +294,7 @@ describe("interactive model setup", () => {
    const prompt = prompter(["", "", ""]);
    prompt.value.ask = async () => {
      asked += 1;
-      if (asked === 4) throw new Error("interrupted");
+      if (asked === 3) throw new Error("interrupted");
      return "";
    };
    const api = clients();
@@ -323,7 +311,7 @@ describe("interactive model setup", () => {
  });

  it("installs selected detected hosts through preview and fingerprint apply", async () => {
-    const prompt = prompter(["", "", "", "n", "", ""]);
+    const prompt = prompter(["", "", "", "", ""]);
    const api = clients();
    const read = vi
      .fn()
@@ -361,7 +349,6 @@ describe("interactive model setup", () => {

    expect(result).toMatchObject({
      state: "configured",
-      model_content_allowed: false,
      installed_hosts: ["codex"],
    });
    expect(read).toHaveBeenNthCalledWith(
@@ -374,7 +361,7 @@ describe("interactive model setup", () => {
  });

  it("lists and installs only hosts that are actually present", async () => {
-    const prompt = prompter(["", "", "", "n", "", ""]);
+    const prompt = prompter(["", "", "", "", ""]);
    const api = clients();
    const read = vi
      .fn()
@@ -421,7 +408,7 @@ describe("interactive model setup", () => {
  });

  it("completes the wizard and keeps onboarding reachable when one host install fails", async () => {
-    const prompt = prompter(["", "", "", "n", "", ""]);
+    const prompt = prompter(["", "", "", "", ""]);
    const api = clients();
    const read = vi
      .fn()
@@ -470,7 +457,6 @@ describe("interactive model setup", () => {
    // calling the install a failure.
    expect(result).toMatchObject({
      state: "configured",
-      model_content_allowed: false,
      installed_hosts: ["codex"],
    });
    const summary = prompt.written.join("\n");
@@ -480,7 +466,7 @@ describe("interactive model setup", () => {
  });

  it("tells the user to install App-only hosts in the App instead of calling them failed", async () => {
-    const prompt = prompter(["", "", "", "n", "", ""]);
+    const prompt = prompter(["", "", "", "", ""]);
    const api = clients();
    const read = vi
      .fn()
@@ -507,7 +493,6 @@ describe("interactive model setup", () => {
    // only the App can install, so nothing counts as "installed".
    expect(result).toMatchObject({
      state: "configured",
-      model_content_allowed: false,
      installed_hosts: [],
    });
    const summary = prompt.written.join("\n");
@@ -16,7 +16,6 @@ import {
  executeModelProfileActivate,
  planModelProfileActivate,
 } from "./model-write-flows.js";
-import { setModelContentPermission } from "./model-content-permission-flow.js";

 type ProviderPreset = {
  id: string;
@@ -122,7 +121,6 @@ export type InteractiveModelSetupState =
  | {
      state: "configured";
      profile: ModelProfile;
-      model_content_allowed: boolean;
      installed_hosts: string[];
    }
  | { state: "skipped" }
@@ -289,15 +287,11 @@ async function installHosts(
 function writeSummary(
  prompter: WizardPrompter,
  profile: ModelProfile,
-  modelContentAllowed: boolean,
  installedHosts: readonly string[],
  failedHosts: readonly { hostId: string; label: string }[],
 ): void {
  prompter.write("配置摘要：");
  prompter.write(`  模型：${profile.name} / ${profile.model}`);
-  prompter.write(
-    `  模型内容外发：${modelContentAllowed ? "已允许" : "已禁止"}`,
-  );
  prompter.write(
    `  应用接入：${installedHosts.length > 0 ? installedHosts.join("、") : "已跳过"}`,
  );
@@ -353,14 +347,6 @@ export async function runInteractiveModelSetup(
      prompter.write("未填写 API Key。输入 back 返回，或重新选择模型服务。");
      continue;
    }
-    const permissionAnswer = (
-      await prompter.ask("允许将对话内容发送给模型以整理记忆？[Y/n]: ")
-    )
-      .trim()
-      .toLowerCase();
-    const modelContentAllowed = !["n", "no", "否", "0"].includes(
-      permissionAnswer,
-    );
    const hostApi = options.hostApi ?? { list: listHosts, read: readHost };
    const selectedHosts = options.installHosts
      ? await chooseDetectedHosts(prompter, options.hostHome, hostApi)
@@ -452,13 +438,6 @@ export async function runInteractiveModelSetup(
      return { state: "failed", code: "model_activate_failed" };
    }

-    const permission = await setModelContentPermission(
-      clients.operations,
-      modelContentAllowed,
-    );
-    if (permission.state !== "ok") {
-      return { state: "failed", code: "model_content_permission_set_failed" };
-    }
    const { installed, failed } = await installHosts(
      options.dataDir,
      selectedHosts,
@@ -467,7 +446,6 @@ export async function runInteractiveModelSetup(
    writeSummary(
      prompter,
      profile,
-      modelContentAllowed,
      installed,
      failed.map((hostId) => ({
        hostId,
@@ -477,7 +455,6 @@ export async function runInteractiveModelSetup(
    return {
      state: "configured",
      profile,
-      model_content_allowed: modelContentAllowed,
      installed_hosts: installed,
    };
  }
@@ -17,6 +17,7 @@ vi.mock("axios", () => ({
  default: { create: vi.fn(() => transport), get: vi.fn() },
 }));

+import type { CoreOperations } from "./core-operations.js";
 import { createCoreOperations } from "./core-operations.js";
 import {
  executeJobAction,
@@ -25,7 +26,6 @@ import {
  planJobAction,
  planJobStart,
 } from "./job-flow.js";
-import type { CoreOperations } from "./core-operations.js";

 function operations(
  post: (path: string, body: Record<string, unknown>) => Promise<unknown>,
@@ -180,5 +180,8 @@ describe("import-job write allow-list", () => {
    await expect(
      operations.post("/api/v1/import-jobs/job-1/delete", {}),
    ).rejects.toThrow("core_operation_path_not_allowed");
+    await expect(
+      operations.post("/api/v1/model-content-permission", { allowed: true }),
+    ).rejects.toThrow("core_operation_path_not_allowed");
  });
 });
@@ -145,9 +145,7 @@ function jobMutationResult(
 export function jobExitCode(document: JobFlowDocument): number {
  if (document.state === "preview" || document.state === "committed") return 0;
  if (document.state === "recovered") return 0;
-  return document.state === "rejected" || document.state === "unknown"
-    ? 1
-    : 2;
+  return document.state === "rejected" || document.state === "unknown" ? 1 : 2;
 }

 function invalid(code: string): JobFlowDocument {
@@ -105,10 +105,7 @@ describe("memhub jobs", () => {
    const output = captureStdout();
    let code: number;
    try {
-      code = await run(
-        ["jobs", "start", "--source", "codex", "--json"],
-        {},
-      );
+      code = await run(["jobs", "start", "--source", "codex", "--json"], {});
    } finally {
      output.restore();
    }
@@ -154,10 +151,7 @@ describe("memhub jobs", () => {
      const output = captureStdout();
      let code: number;
      try {
-        code = await run(
-          ["jobs", action, "--job", "job-1", "--json"],
-          {},
-        );
+        code = await run(["jobs", action, "--job", "job-1", "--json"], {});
      } finally {
        output.restore();
      }
@@ -1,7 +1,10 @@
-import type { DiagnosticEventV1 } from "@one-memory/runtime-control";
-import { diagnosticLogFiles, readDiagnosticEvents } from "@one-memory/runtime-control";
 import { lstat } from "node:fs/promises";
 import path from "node:path";
+import type { DiagnosticEventV1 } from "@one-memory/runtime-control";
+import {
+  diagnosticLogFiles,
+  readDiagnosticEvents,
+} from "@one-memory/runtime-control";

 export interface LogQuery {
  since?: Date;
@@ -160,10 +163,21 @@ export async function readLogReport(
  const logsDirectory = path.join(path.resolve(dataDir), "logs");
  const info = await lstat(logsDirectory).catch(() => null);
  if (info !== null && (!info.isDirectory() || info.isSymbolicLink())) {
-    return { events: [], invalid_lines: 0, failure: "log_directory_unreadable" };
+    return {
+      events: [],
+      invalid_lines: 0,
+      failure: "log_directory_unreadable",
+    };
  }
-  if (info !== null && (await diagnosticLogFiles(dataDir).catch(() => null)) === null) {
-    return { events: [], invalid_lines: 0, failure: "log_directory_unreadable" };
+  if (
+    info !== null &&
+    (await diagnosticLogFiles(dataDir).catch(() => null)) === null
+  ) {
+    return {
+      events: [],
+      invalid_lines: 0,
+      failure: "log_directory_unreadable",
+    };
  }
  const explicit =
    query.since ||
@@ -239,15 +253,27 @@ export function redactDiagnosticEvent(
    outcome: event.outcome,
    session_id: event.session_id,
    ...(event.trace_id === undefined ? {} : { trace_id: event.trace_id }),
-    ...(event.incident_id === undefined ? {} : { incident_id: event.incident_id }),
-    ...(event.instance_id === undefined ? {} : { instance_id: event.instance_id }),
-    ...(event.operation_id === undefined ? {} : { operation_id: event.operation_id }),
+    ...(event.incident_id === undefined
+      ? {}
+      : { incident_id: event.incident_id }),
+    ...(event.instance_id === undefined
+      ? {}
+      : { instance_id: event.instance_id }),
+    ...(event.operation_id === undefined
+      ? {}
+      : { operation_id: event.operation_id }),
    ...(event.request_id === undefined ? {} : { request_id: event.request_id }),
    ...(event.job_id === undefined ? {} : { job_id: event.job_id }),
    ...(event.task_id === undefined ? {} : { task_id: event.task_id }),
-    ...(event.duration_ms === undefined ? {} : { duration_ms: event.duration_ms }),
-    ...(event.state_before === undefined ? {} : { state_before: event.state_before }),
-    ...(event.state_after === undefined ? {} : { state_after: event.state_after }),
+    ...(event.duration_ms === undefined
+      ? {}
+      : { duration_ms: event.duration_ms }),
+    ...(event.state_before === undefined
+      ? {}
+      : { state_before: event.state_before }),
+    ...(event.state_after === undefined
+      ? {}
+      : { state_after: event.state_after }),
    ...(event.error_code === undefined ? {} : { error_code: event.error_code }),
    ...(event.error_type === undefined ? {} : { error_type: event.error_type }),
    ...(event.safe_stack === undefined
@@ -11,9 +11,8 @@ import {
 } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
-import { describe, expect, it } from "vitest";
-
 import type { CliLinkPlan } from "@one-memory/runtime-control";
+import { describe, expect, it } from "vitest";
 import {
  installManagedCliAlias,
  installManagedCliLink,
@@ -107,7 +106,9 @@ describe("managed memhub CLI alias", () => {
    await installManagedCliAlias(plan);

    await expect(removeManagedCliAlias(plan)).resolves.toBe("removed");
-    await expect(lstat(plan.link_path)).rejects.toMatchObject({ code: "ENOENT" });
+    await expect(lstat(plan.link_path)).rejects.toMatchObject({
+      code: "ENOENT",
+    });
  });

  // C8 P1 (2026-09-29): `runtime-control.applyLocalUninstall` deletes
@@ -156,7 +157,10 @@ describe("managed memhub CLI alias", () => {
  it("manages the one-memory companion when memhub is already present", async () => {
    if (process.platform === "win32") return;
    const f = await fixture();
-    const primary = { ...f.base, link_path: path.join(f.base.bin_directory, "memhub") };
+    const primary = {
+      ...f.base,
+      link_path: path.join(f.base.bin_directory, "memhub"),
+    };
    await mkdir(f.base.bin_directory, { recursive: true, mode: 0o700 });
    await (await import("node:fs/promises")).chmod(f.base.bin_directory, 0o700);
    await (await import("node:fs/promises")).symlink(
@@ -168,7 +172,10 @@ describe("managed memhub CLI alias", () => {
    const plan = await planManagedCliLink(f.base, "one-memory");
    // The companion does not exist yet, so it is ours to create; crucially it
    // targets the compat name, not the package's `memhub`.
-    expect(plan).toMatchObject({ status: "create", link_path: f.base.link_path });
+    expect(plan).toMatchObject({
+      status: "create",
+      link_path: f.base.link_path,
+    });

    // The package's memhub link is untouched.
    expect(await readlink(primary.link_path)).toBe(f.target);
@@ -1,4 +1,12 @@
-import { access, chmod, lstat, mkdir, readlink, rm, symlink } from "node:fs/promises";
+import {
+  access,
+  chmod,
+  lstat,
+  mkdir,
+  readlink,
+  rm,
+  symlink,
+} from "node:fs/promises";
 import path from "node:path";

 import type { CliLinkPlan, CliLinkReceipt } from "@one-memory/runtime-control";
@@ -1,152 +0,0 @@
-import { beforeEach, describe, expect, it, vi } from "vitest";
-
-import { run } from "./cli.js";
-
-const transport = vi.hoisted(() => ({
-  get: vi.fn(),
-  post: vi.fn(),
-  put: vi.fn(),
-  patch: vi.fn(),
-  delete: vi.fn(),
-}));
-vi.mock("axios", () => ({
-  default: { create: vi.fn(() => transport), get: vi.fn() },
-}));
-
-const runtime = vi.hoisted(() => ({
-  inspectRuntimeStatus: vi.fn(),
-  discoverRuntimeAttachment: vi.fn(),
-}));
-vi.mock("@one-memory/runtime-control", async (importOriginal) => ({
-  ...(await importOriginal<typeof import("@one-memory/runtime-control")>()),
-  ...runtime,
-}));
-
-beforeEach(() => {
-  vi.clearAllMocks();
-  runtime.inspectRuntimeStatus.mockResolvedValue({
-    state: "running",
-    code: "runtime_verified",
-    instanceId: "instance-1",
-    host: "127.0.0.1",
-    port: 8765,
-    version: "1.2.3",
-  });
-  runtime.discoverRuntimeAttachment.mockResolvedValue({
-    instanceId: "instance-1",
-    apiToken: "synthetic-runtime-token",
-    host: "127.0.0.1",
-    port: 8765,
-    servicePid: 4242,
-    health: {
-      status: "ok",
-      service: "one-memory-core",
-      version: "1.2.3",
-      instance_id: "instance-1",
-    },
-  });
-});
-
-describe("model content permission CLI", () => {
-  it("documents only final-state permission and manual extraction commands", async () => {
-    const permission = await capturedRun(["help", "permission"]);
-    expect(permission.code).toBe(0);
-    expect(permission.stdout).toContain("memhub permission allow");
-    expect(permission.stdout).toContain("memhub permission deny");
-    expect(permission.stdout).not.toMatch(/\b(preview|confirm|process|revoke)\b/);
-    expect(permission.stdout).not.toContain("revision");
-    expect(permission.stdout).not.toContain("receipt");
-
-    const processing = await capturedRun(["help", "processing"]);
-    expect(processing.code).toBe(0);
-    expect(processing.stdout).toContain("memhub processing extract");
-    expect(processing.stdout).not.toMatch(/\b(plan|apply|receipt)\b/);
-    expect(processing.stdout).not.toContain("backlog");
-  });
-
-  it("reads the final state", async () => {
-    transport.get.mockResolvedValue({ status: 200, data: { allowed: true } });
-
-    const result = await capturedRun(["permission", "status", "--json"]);
-
-    expect(result.code).toBe(0);
-    expect(JSON.parse(result.stdout)).toMatchObject({
-      state: "ok",
-      allowed: true,
-    });
-    expect(transport.get).toHaveBeenCalledWith(
-      "/api/v1/model-content-permission",
-      undefined,
-    );
-  });
-
-  it.each([
-    ["allow", true],
-    ["deny", false],
-  ] as const)("sets %s directly with only allowed", async (verb, allowed) => {
-    transport.post.mockResolvedValue({ status: 200, data: { allowed } });
-
-    const result = await capturedRun(["permission", verb, "--json"]);
-
-    expect(result.code).toBe(0);
-    expect(JSON.parse(result.stdout)).toMatchObject({ state: "ok", allowed });
-    expect(transport.post).toHaveBeenCalledWith(
-      "/api/v1/model-content-permission",
-      { allowed },
-      undefined,
-    );
-  });
-
-  it.each(["preview", "confirm", "process", "revoke"])(
-    "rejects removed %s commands before runtime access",
-    async (verb) => {
-      const result = await capturedRun(["permission", verb, "--json"]);
-
-      expect(result.code).toBe(2);
-      expect(runtime.inspectRuntimeStatus).not.toHaveBeenCalled();
-      expect(transport.get).not.toHaveBeenCalled();
-      expect(transport.post).not.toHaveBeenCalled();
-    },
-  );
-
-  it.each(["plan", "apply", "receipt"])(
-    "rejects removed processing %s commands before runtime access",
-    async (verb) => {
-      const result = await capturedRun(["processing", verb, "--json"]);
-
-      expect(result.code).toBe(2);
-      expect(runtime.inspectRuntimeStatus).not.toHaveBeenCalled();
-      expect(transport.get).not.toHaveBeenCalled();
-      expect(transport.post).not.toHaveBeenCalled();
-    },
-  );
-});
-
-async function capturedRun(arguments_: string[]) {
-  let stdout = "";
-  let stderr = "";
-  const stdoutSpy = vi
-    .spyOn(process.stdout, "write")
-    .mockImplementation((chunk) => {
-      stdout += String(chunk);
-      return true;
-    });
-  const stderrSpy = vi
-    .spyOn(process.stderr, "write")
-    .mockImplementation((chunk) => {
-      stderr += String(chunk);
-      return true;
-    });
-  try {
-    return {
-      code: await run(arguments_, {
-        ONE_MEMORY_DATA_DIR: "/synthetic/data",
-      }),
-      stdout,
-      stderr,
-    };
-  } finally {
-    stdoutSpy.mockRestore();
-    stderrSpy.mockRestore();
-  }
-}
@@ -1,63 +0,0 @@
-import { describe, expect, it, vi } from "vitest";
-
-import type { CoreReader } from "./core-client.js";
-import type { CoreOperations } from "./core-operations.js";
-import {
-  readModelContentPermission,
-  setModelContentPermission,
-} from "./model-content-permission-flow.js";
-
-describe("model content permission flow", () => {
-  it("reads the single final-state boolean", async () => {
-    const reader = vi.fn().mockResolvedValue({
-      state: "ok",
-      data: { allowed: true },
-    });
-
-    await expect(
-      readModelContentPermission(reader as CoreReader),
-    ).resolves.toMatchObject({ state: "ok", allowed: true });
-    expect(reader).toHaveBeenCalledWith("/api/v1/model-content-permission");
-  });
-
-  it.each([
-    ["allow", true],
-    ["deny", false],
-  ] as const)("sets %s directly without revision or receipt fields", async (_, allowed) => {
-    const post = vi.fn().mockResolvedValue({
-      state: "committed",
-      data: { allowed },
-    });
-
-    await expect(
-      setModelContentPermission({ post } as CoreOperations, allowed),
-    ).resolves.toMatchObject({ state: "ok", allowed });
-    expect(post).toHaveBeenCalledWith("/api/v1/model-content-permission", {
-      allowed,
-    });
-  });
-
-  it("does not accept a malformed or mismatched success response", async () => {
-    const malformed = vi.fn().mockResolvedValue({
-      state: "committed",
-      data: { allowed: true, revision: 1 },
-    });
-    await expect(
-      setModelContentPermission({ post: malformed } as CoreOperations, true),
-    ).resolves.toMatchObject({
-      state: "unknown",
-      code: "model_content_permission_response_invalid",
-    });
-
-    const mismatched = vi.fn().mockResolvedValue({
-      state: "committed",
-      data: { allowed: false },
-    });
-    await expect(
-      setModelContentPermission({ post: mismatched } as CoreOperations, true),
-    ).resolves.toMatchObject({
-      state: "unknown",
-      code: "model_content_permission_state_mismatch",
-    });
-  });
-});
@@ -1,93 +0,0 @@
-import type { CoreReader } from "./core-client.js";
-import type { CoreOperations } from "./core-operations.js";
-import { PRODUCT } from "./version.js";
-
-export type ModelContentPermissionDocument =
-  | {
-      schema_version: 1;
-      product: string;
-      state: "ok";
-      allowed: boolean;
-    }
-  | {
-      schema_version: 1;
-      product: string;
-      state: "rejected" | "unknown";
-      code: string;
-    };
-
-export async function readModelContentPermission(
-  reader: CoreReader,
-): Promise<ModelContentPermissionDocument> {
-  const outcome = await reader<unknown>("/api/v1/model-content-permission");
-  if (outcome.state !== "ok") {
-    return failure("unknown", outcome.code);
-  }
-  return permissionState(outcome.data, "model_content_permission_response_invalid");
-}
-
-export async function setModelContentPermission(
-  operations: CoreOperations,
-  allowed: boolean,
-): Promise<ModelContentPermissionDocument> {
-  const outcome = await operations.post<unknown>(
-    "/api/v1/model-content-permission",
-    { allowed },
-  );
-  if (outcome.state === "rejected") {
-    return failure("rejected", outcome.code);
-  }
-  if (outcome.state === "unknown") {
-    return failure("unknown", outcome.code);
-  }
-  const document = permissionState(
-    outcome.data,
-    "model_content_permission_response_invalid",
-  );
-  return document.state === "ok" && document.allowed !== allowed
-    ? failure("unknown", "model_content_permission_state_mismatch")
-    : document;
-}
-
-export function modelContentPermissionExitCode(
-  document: ModelContentPermissionDocument,
-): number {
-  return document.state === "ok" ? 0 : 1;
-}
-
-export function renderModelContentPermissionText(
-  document: ModelContentPermissionDocument,
-): string {
-  if (document.state === "ok") {
-    return `模型内容外发：${document.allowed ? "允许" : "禁止"}\n`;
-  }
-  return `模型内容权限未更新：${document.state}（${document.code}）\n`;
-}
-
-function permissionState(
-  value: unknown,
-  invalidCode: string,
-): ModelContentPermissionDocument {
-  if (
-    value === null ||
-    typeof value !== "object" ||
-    Array.isArray(value) ||
-    Object.keys(value).length !== 1 ||
-    typeof (value as Record<string, unknown>).allowed !== "boolean"
-  ) {
-    return failure("unknown", invalidCode);
-  }
-  return {
-    schema_version: 1,
-    product: PRODUCT,
-    state: "ok",
-    allowed: (value as { allowed: boolean }).allowed,
-  };
-}
-
-function failure(
-  state: "rejected" | "unknown",
-  code: string,
-): ModelContentPermissionDocument {
-  return { schema_version: 1, product: PRODUCT, state, code };
-}
@@ -13,8 +13,8 @@
 import { createHash, randomUUID } from "node:crypto";
 import type { ModelProfile } from "@one-memory/contracts";
 import {
-  ModelProfileInputError,
  buildModelProfileUpdateRequest,
+  ModelProfileInputError,
  parseModelProfileReceipt,
 } from "@one-memory/operations";
 import type { CoreOperations, CorePatchOperations } from "./core-operations.js";
@@ -78,7 +78,10 @@ function digestOf(value: unknown): string {

 function preview(
  plan:
-    | Omit<Extract<ModelWritePlan, { kind: "model_profile_activate" }>, "digest">
+    | Omit<
+        Extract<ModelWritePlan, { kind: "model_profile_activate" }>,
+        "digest"
+      >
    | Omit<Extract<ModelWritePlan, { kind: "model_profile_edit" }>, "digest">,
 ): ModelWriteDocument {
  return {
@@ -198,9 +201,9 @@ export async function executeModelProfileEdit(
  let request: ReturnType<typeof buildModelProfileUpdateRequest>;
  try {
    request = buildModelProfileUpdateRequest(plan.profile_id, {
-    ...plan.patch,
-    expected_revision: plan.expected_revision,
-    ...(apiKey === undefined ? {} : { api_key: apiKey }),
+      ...plan.patch,
+      expected_revision: plan.expected_revision,
+      ...(apiKey === undefined ? {} : { api_key: apiKey }),
    });
  } catch (error) {
    if (error instanceof ModelProfileInputError) {
@@ -1,7 +1,7 @@
 import { execFile } from "node:child_process";
 import { createHash } from "node:crypto";
 import { mkdtemp, readFile, writeFile } from "node:fs/promises";
-import { tmpdir, homedir } from "node:os";
+import { homedir, tmpdir } from "node:os";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
 import { promisify } from "node:util";
@@ -137,8 +137,8 @@ describe("package matrix harness", () => {
    };
    expect(document.state).toBe("verified");
    const results = document.commands.stopped_destructive.results;
-    // 13 destructive commands, each unknown/runtime_not_running with exit 3.
-    expect(Object.keys(results)).toHaveLength(13);
+    // 11 destructive commands, each unknown/runtime_not_running with exit 3.
+    expect(Object.keys(results)).toHaveLength(11);
    for (const [name, result] of Object.entries(results)) {
      expect(result, `${name} did not fail closed`).toEqual({
        exit: 3,
@@ -162,7 +162,7 @@ describe("package matrix harness", () => {
    const shim = [
      "#!/usr/bin/env node",
      "const a=process.argv.slice(2);",
-      "if(['calls','import-file','notes','permission','extraction','graph','models'].includes(a[0])||(a[0]==='processing'&&a[1]==='extract')){",
+      "if(['calls','import-file','notes','extraction','graph','models'].includes(a[0])||(a[0]==='processing'&&a[1]==='extract')){",
      '  process.stdout.write(\'{"schema_version":1,"product":"memhub","state":"committed","operation_id":"x","receipt":{}}\\n\');',
      "  process.exit(0);",
      "}",
@@ -188,7 +188,7 @@ describe("package matrix harness", () => {
    expect(failure?.state).toBe("failed");
    expect(
      failure?.failures.filter((item) => item.includes("did not fail closed")),
-    ).toHaveLength(13);
+    ).toHaveLength(11);
  });

  it("can require an expected bundle SHA and reports bundle metadata", {
@@ -280,9 +280,8 @@ describe("package matrix harness", () => {
  });

  it("rejects a catalog that declares the parent command but not the new verb", async () => {
-    // A bundle can still declare `permission`/`models` while missing
-    // `permission allow` or `models activate`. Names alone would pass it; the
-    // usage marker must not.
+    // A bundle can still declare `models` while missing `models activate`.
+    // Names alone would pass it; the usage marker must not.
    const declared = [
      "install",
      "init",
@@ -292,7 +291,6 @@ describe("package matrix harness", () => {
      "calls",
      "import-file",
      "notes",
-      "permission",
      "extraction",
      "graph",
      "models",
@@ -342,9 +340,6 @@ describe("package matrix harness", () => {
      },
    };
    const failures = harness.validateCatalog(catalog);
-    expect(failures).toContain(
-      'command "permission" is missing usage "memhub permission allow"',
-    );
    expect(failures).toContain(
      'command "models" is missing usage "memhub models activate"',
    );
@@ -1,6 +1,6 @@
 import { execFileSync } from "node:child_process";
 import { existsSync } from "node:fs";
-import { mkdtemp, readFile, readdir, rm } from "node:fs/promises";
+import { mkdtemp, readdir, readFile, rm } from "node:fs/promises";
 import os from "node:os";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
@@ -151,9 +151,6 @@ describe("memhub short-command contract", () => {
      "memhub import-file discard",
      "memhub notes add",
      "memhub notes process",
-      "memhub permission status",
-      "memhub permission allow",
-      "memhub permission deny",
      "memhub models setup",
      "memhub models activate",
      "memhub models edit",
@@ -163,6 +160,9 @@ describe("memhub short-command contract", () => {
    ]) {
      expect(flattened).toContain(shortForm);
    }
+    expect(COMMANDS.some((command) => command.name === "permission")).toBe(
+      false,
+    );
  });

  it("ships a memhub bin entry pointing at a real bundle path", () => {
@@ -103,26 +103,45 @@ describe("unknown flags are rejected", () => {
  });

  it("still accepts the documented verbless scan", () => {
-    expect(parseProductCommand("scan", ["--all", "--yes"])?.plan.all).toBe(true);
-    expect(parseProductCommand("scan", ["--source", "codex"])?.plan.sourceId).toBe(
-      "codex",
+    expect(parseProductCommand("scan", ["--all", "--yes"])?.plan.all).toBe(
+      true,
    );
+    expect(
+      parseProductCommand("scan", ["--source", "codex"])?.plan.sourceId,
+    ).toBe("codex");
  });

  it("rejects flags that the named verb does not consume", () => {
    // These were accepted and then silently dropped. An operator who typed
    // `path remove --source codex` would believe the source filter applied.
    expect(
-      parseProductCommand("path", ["remove", "--root", "/tmp/x", "--source", "codex"]),
+      parseProductCommand("path", [
+        "remove",
+        "--root",
+        "/tmp/x",
+        "--source",
+        "codex",
+      ]),
    ).toBeNull();
    expect(
-      parseProductCommand("path", ["remove", "--root", "/tmp/x", "--no-recursive"]),
+      parseProductCommand("path", [
+        "remove",
+        "--root",
+        "/tmp/x",
+        "--no-recursive",
+      ]),
    ).toBeNull();
    expect(
      parseProductCommand("path", ["add", "--root", "/tmp/x", "--full"]),
    ).toBeNull();
    expect(
-      parseProductCommand("backup", ["create", "--target", "/tmp/b", "--source", "/tmp/s"]),
+      parseProductCommand("backup", [
+        "create",
+        "--target",
+        "/tmp/b",
+        "--source",
+        "/tmp/s",
+      ]),
    ).toBeNull();
    expect(
      parseProductCommand("backup", [
@@ -163,9 +182,7 @@ describe("unknown flags are rejected", () => {
  });

  it("rejects repeated boolean flags instead of treating them as idempotent", () => {
-    expect(
-      parseProductCommand("scan", ["--all", "--all", "--yes"]),
-    ).toBeNull();
+    expect(parseProductCommand("scan", ["--all", "--all", "--yes"])).toBeNull();
    expect(
      parseProductCommand("scan", ["--full", "--full", "--yes"]),
    ).toBeNull();
@@ -179,7 +196,13 @@ describe("unknown flags are rejected", () => {
      ]),
    ).toBeNull();
    expect(
-      parseProductCommand("backup", ["create", "--target", "/tmp/a", "--yes", "--yes"]),
+      parseProductCommand("backup", [
+        "create",
+        "--target",
+        "/tmp/a",
+        "--yes",
+        "--yes",
+      ]),
    ).toBeNull();
  });

@@ -215,12 +238,12 @@ describe("unknown flags are rejected", () => {
    ]);
    expect(parsed).not.toBeNull();
    expect(parsed?.confirmPath).toBe("/tmp/backups/a.sqlite3");
-    expect(restoreConfirmationMatches("/tmp/backups/a.sqlite3", parsed?.confirmPath)).toBe(
-      true,
-    );
-    expect(restoreConfirmationMatches("/tmp/backups/a.sqlite3", undefined)).toBe(
-      false,
-    );
+    expect(
+      restoreConfirmationMatches("/tmp/backups/a.sqlite3", parsed?.confirmPath),
+    ).toBe(true);
+    expect(
+      restoreConfirmationMatches("/tmp/backups/a.sqlite3", undefined),
+    ).toBe(false);
    expect(
      restoreConfirmationMatches(
        "/tmp/backups/a.sqlite3",
@@ -365,7 +388,11 @@ describe("path add and remove carry what the endpoint needs", () => {
  });

  it("distinguishes remove from add", () => {
-    const parsed = parseProductCommand("path", ["remove", "--root", "/tmp/scanme"]);
+    const parsed = parseProductCommand("path", [
+      "remove",
+      "--root",
+      "/tmp/scanme",
+    ]);
    if (parsed === null) {
      throw new Error("path remove must parse");
    }
@@ -375,7 +402,12 @@ describe("path add and remove carry what the endpoint needs", () => {
  });

  it("does not silently drop --yes", () => {
-    const parsed = parseProductCommand("path", ["add", "--root", "/tmp/a", "--yes"]);
+    const parsed = parseProductCommand("path", [
+      "add",
+      "--root",
+      "/tmp/a",
+      "--yes",
+    ]);
    expect(parsed?.confirmed).toBe(true);
    // Without this, a typo'd `--yes` would fall through to a plan and the user
    // would believe the change had been made.
@@ -6,10 +6,7 @@ import { fileURLToPath } from "node:url";
 import { describe, expect, it } from "vitest";

 /** The app root (one level above this test's `src/`). */
-const root = path.resolve(
-  path.dirname(fileURLToPath(import.meta.url)),
-  "..",
-);
+const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
 const bundle = path.join(root, "dist", "one-memory.cjs");

 // The published npm package must be self-contained. The bundle inlines
@@ -20,9 +17,10 @@ const bundle = path.join(root, "dist", "one-memory.cjs");
 describe("publish manifest is production-safe", () => {
  it("declares no npm runtime dependency once dependencies are inlined", () => {
    // `test` builds with tsup first, so a missing bundle is a real failure.
-    expect(existsSync(bundle), "run `pnpm --filter @one-memory/cli build`").toBe(
-      true,
-    );
+    expect(
+      existsSync(bundle),
+      "run `pnpm --filter @one-memory/cli build`",
+    ).toBe(true);
    const source = readFileSync(bundle, "utf8");
    const external = [
      ...new Set(
@@ -37,16 +35,15 @@ describe("publish manifest is production-safe", () => {
      ),
    ];
    // Every one of these must be a Node builtin, i.e. no npm package.
-    const npmDeps = external.filter(
-      (name) => !builtinModules.includes(name),
-    );
+    const npmDeps = external.filter((name) => !builtinModules.includes(name));
    expect(npmDeps).toEqual([]);
  });

  it("strip-out rule keeps only what the bundle requires", () => {
-    expect(existsSync(bundle), "run `pnpm --filter @one-memory/cli build`").toBe(
-      true,
-    );
+    expect(
+      existsSync(bundle),
+      "run `pnpm --filter @one-memory/cli build`",
+    ).toBe(true);
    // Run the real generator, then inspect its output manifest.
    execFileSync("node", ["scripts/prepare-publish-manifest.mjs"], {
      cwd: root,
@@ -840,6 +840,7 @@ describe("CLI-21 memory reads", () => {
    expect(isReadOnlyPath("/api/v1/memories/graph")).toBe(false);
    expect(isReadOnlyPath("/api/v1/memories/ask")).toBe(false);
    expect(isReadOnlyPath("/api/v1/memories/m1")).toBe(true);
+    expect(isReadOnlyPath("/api/v1/model-content-permission")).toBe(false);
    expect(
      isReadOnlyPath(
        "/api/v1/memory-mutations/123e4567-e89b-12d3-a456-426614174000",
@@ -49,9 +49,9 @@ describe("runtime service lock guard", () => {
      JSON.stringify({ pid: 4321 }),
      { mode: 0o600 },
    );
-    await expect(
-      inspectRuntimeServiceLock(dataDir, () => false),
-    ).resolves.toBe("clear");
+    await expect(inspectRuntimeServiceLock(dataDir, () => false)).resolves.toBe(
+      "clear",
+    );
  });

  it("fails closed on a malformed lease", async () => {
@@ -1,15 +1,15 @@
-import { mkdtemp, mkdir, writeFile } from "node:fs/promises";
+import { mkdir, mkdtemp, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path from "node:path";
 import { describe, expect, it, vi } from "vitest";

 import {
+  type CommandResult,
  inspectLaunchdService,
  restartManagedService,
+  type ServiceCommandRunner,
  startManagedService,
  stopManagedService,
-  type CommandResult,
-  type ServiceCommandRunner,
 } from "./service-session.js";

 const SERVICE_ID = "com.tencent.onememory.runtime";
@@ -35,7 +35,11 @@ describe("service-aware launchd session", () => {
  it("reports running with the launchd pid and plist", async () => {
    const run = vi.fn<ServiceCommandRunner>(async () => running(84237));
    await expect(
-      inspectLaunchdService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      inspectLaunchdService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toEqual({
      state: "running",
      pid: 84237,
@@ -46,7 +50,11 @@ describe("service-aware launchd session", () => {
  it("treats 'could not find service' as absent", async () => {
    const run = vi.fn<ServiceCommandRunner>(async () => absent());
    await expect(
-      inspectLaunchdService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      inspectLaunchdService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toEqual({ state: "absent" });
  });

@@ -57,7 +65,11 @@ describe("service-aware launchd session", () => {
      stderr: "Input/output error",
    }));
    await expect(
-      inspectLaunchdService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      inspectLaunchdService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toEqual({ state: "unknown", code: "launchd_inspection_failed" });
  });

@@ -146,13 +158,21 @@ describe("service-aware launchd session", () => {
      if (command[1] === "print") {
        printed += 1;
        return printed === 1
-          ? { code: 0, stdout: `path = /tmp/x.plist\nstate = not running\n`, stderr: "" }
+          ? {
+              code: 0,
+              stdout: `path = /tmp/x.plist\nstate = not running\n`,
+              stderr: "",
+            }
          : running(84237);
      }
      return running(84237);
    });
    await expect(
-      startManagedService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      startManagedService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toBe("started");
    expect(calls).toContainEqual([
      "launchctl",
@@ -169,7 +189,11 @@ describe("service-aware launchd session", () => {
      return running(84237);
    });
    await expect(
-      startManagedService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      startManagedService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toBe("started");
    // Already running -> no kickstart needed.
    expect(calls.some((c) => c[1] === "kickstart")).toBe(false);
@@ -214,7 +238,11 @@ describe("service-aware launchd session", () => {
      stderr: "",
    }));
    await expect(
-      inspectLaunchdService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      inspectLaunchdService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toEqual({ state: "loaded", plistPath: "/tmp/x.plist" });
  });

@@ -225,7 +253,11 @@ describe("service-aware launchd session", () => {
      return running(101);
    });
    await expect(
-      restartManagedService({ serviceId: SERVICE_ID, uid: UID, runCaptured: run }),
+      restartManagedService({
+        serviceId: SERVICE_ID,
+        uid: UID,
+        runCaptured: run,
+      }),
    ).resolves.toBe("restarted");
    expect(calls).toContainEqual([
      "launchctl",
@@ -286,7 +318,9 @@ describe("service-aware launchd session", () => {
        stdout: JSON.stringify({
          Label: SERVICE_ID,
          ProgramArguments: ["/tmp/node", "/tmp/runtime.cjs", "serve"],
-          EnvironmentVariables: { ONE_MEMORY_DATA_DIR: "/tmp/this-install/data" },
+          EnvironmentVariables: {
+            ONE_MEMORY_DATA_DIR: "/tmp/this-install/data",
+          },
        }),
        stderr: "",
      };
@@ -49,7 +49,10 @@ function launchdDomain(uid: number): string {
 /** Deterministic location of the managed LaunchAgent plist. Needed because
 * after `bootout` launchd no longer reports a path, yet `start` must be able to
 * bootstrap the very same service back. */
-export function defaultLaunchAgentPath(home: string, serviceId: string): string {
+export function defaultLaunchAgentPath(
+  home: string,
+  serviceId: string,
+): string {
  return path.posix.join(home, "Library", "LaunchAgents", `${serviceId}.plist`);
 }

@@ -162,7 +165,10 @@ export async function stopManagedService(options: {
    ? (["launchctl", "bootout", domain, before.plistPath] as const)
    : (["launchctl", "bootout", `${domain}/${options.serviceId}`] as const);
  const result = await options.runCaptured([...target]);
-  if (result.code !== 0 && !/could not find service|service not found/i.test(result.stderr)) {
+  if (
+    result.code !== 0 &&
+    !/could not find service|service not found/i.test(result.stderr)
+  ) {
    throw new Error("launchd_bootout_failed");
  }
  // "Stopped" must mean the label stays gone, not that we saw it gone once.
@@ -295,9 +301,7 @@ export async function runLaunchctlCaptured(
      stderr += String(chunk);
    });
    child.once("error", () => reject(new Error("launchd_command_failed")));
-    child.once("close", (code) =>
-      resolve({ code: code ?? 1, stdout, stderr }),
-    );
+    child.once("close", (code) => resolve({ code: code ?? 1, stdout, stderr }));
  });
 }

@@ -3,7 +3,11 @@ import { tmpdir } from "node:os";
 import path from "node:path";
 import { describe, expect, it } from "vitest";

-import { applyShellPath, planShellPath, removeShellPath } from "./shell-path.js";
+import {
+  applyShellPath,
+  planShellPath,
+  removeShellPath,
+} from "./shell-path.js";

 async function home(): Promise<string> {
  return mkdtemp(path.join(tmpdir(), "memhub-shell-path-"));
@@ -72,9 +72,7 @@ function blockFor(binDirectory: string): string {
 * original file byte-for-byte regardless of the original's trailing newline. */
 function ownedBlock(existing: string | null, binDirectory: string): string {
  const separator =
-    existing === null || existing === "" || existing.endsWith("\n")
-      ? ""
-      : "\n";
+    existing === null || existing === "" || existing.endsWith("\n") ? "" : "\n";
  return `${separator}${blockFor(binDirectory)}`;
 }

@@ -104,10 +102,7 @@ export async function planShellPath(options: {
      disabled: true,
    };
  }
-  const blocked = (
-    reason: string,
-    target: string,
-  ): ShellPathPlan => ({
+  const blocked = (reason: string, target: string): ShellPathPlan => ({
    schema_version: 1,
    status: "blocked",
    blocked_reason: reason,
@@ -229,11 +224,12 @@ async function writeProfileAtomically(
  // A fresh `--home` may not exist yet; create it (owner-only) before writing
  // the temp profile, otherwise the write fails with a raw ENOENT. Surface a
  // structured code for any other failure.
-  await mkdir(path.dirname(profilePath), { recursive: true, mode: 0o700 }).catch(
-    () => {
-      throw new Error("shell_path_home_unwritable");
-    },
-  );
+  await mkdir(path.dirname(profilePath), {
+    recursive: true,
+    mode: 0o700,
+  }).catch(() => {
+    throw new Error("shell_path_home_unwritable");
+  });
  const temporary = `${profilePath}.memhub.${process.pid}.tmp`;
  try {
    await writeFile(temporary, content, { encoding: "utf8", mode: 0o644 });
@@ -468,7 +468,9 @@ describe("CLI-19 progress and issue reads", () => {
    );
    expect(output).toContain("[15:04:05.067] progress #1");
    expect(output).toContain("[15:04:05.067] progress #2");
-    expect(output.match(/\x1b\[2J\x1b\[H/g)).toHaveLength(2);
+    expect(output.match(new RegExp("\\x1b\\[2J\\x1b\\[H", "g"))).toHaveLength(
+      2,
+    );
  });

  it("keeps non-terminal watch output append-only", async () => {
@@ -1,9 +1,6 @@
 import { describe, expect, it, vi } from "vitest";
 import type { CoreReader, ReadOutcome } from "./core-client.js";
-import type {
-  CoreOperations,
-  OperationOutcome,
-} from "./core-operations.js";
+import type { CoreOperations, OperationOutcome } from "./core-operations.js";
 import {
  renderSourceSyncText,
  setSourceSync,
@@ -320,4 +317,3 @@ describe("setSourceSync", () => {
    expect(renderSourceSyncText(invalid)).toBe("未变更：invalid_source_id\n");
  });
 });
-
@@ -1,7 +1,7 @@
 import {
-  SourceSyncInputError,
  buildSourceSyncRequest,
  hasSourceSyncTarget,
+  SourceSyncInputError,
 } from "@one-memory/operations";
 import type { CoreReader } from "./core-client.js";
 import type { CoreOperations } from "./core-operations.js";
@@ -36,7 +36,10 @@ async function current(
  reader: CoreReader,
  sourceId: string,
 ): Promise<
-  | { state: "ok"; source: NonNullable<ReturnType<typeof validateSources>>[number] }
+  | {
+      state: "ok";
+      source: NonNullable<ReturnType<typeof validateSources>>[number];
+    }
  | { state: "invalid" | "unknown"; code: string }
 > {
  const outcome = await reader<unknown>("/api/v1/sources");
@@ -30,6 +30,8 @@ describe("status default output", () => {
    expect(text).not.toContain("[object Object]");
    expect(text).toContain("卡住 类型 整理对话/原因 等待模型/数量 2");
    expect(text).not.toMatch(/background_verified|additions_only|error_codes=/);
-    expect(text).not.toMatch(/provider_kind|has_api_key|credential_state|observed_at/);
+    expect(text).not.toMatch(
+      /provider_kind|has_api_key|credential_state|observed_at/,
+    );
  });
 });
@@ -85,12 +85,16 @@ describe("memhub ui command wiring", () => {
    const output = capture();
    let code: number;
    try {
-      code = await run(["ui", "--json"], {}, {
-        openBrowser: async (url) => {
-          opened.push(url);
-          return true;
+      code = await run(
+        ["ui", "--json"],
+        {},
+        {
+          openBrowser: async (url) => {
+            opened.push(url);
+            return true;
+          },
        },
-      });
+      );
    } finally {
      output.restore();
    }
@@ -102,7 +106,9 @@ describe("memhub ui command wiring", () => {
      state: "opened",
      url: "http://127.0.0.1:8765/ui/login?code=one-time-code",
    });
-    expect(opened).toEqual(["http://127.0.0.1:8765/ui/login?code=one-time-code"]);
+    expect(opened).toEqual([
+      "http://127.0.0.1:8765/ui/login?code=one-time-code",
+    ]);
    // The bearer token was sent to Core in the request header, never in output.
    expect(output.value()).not.toMatch(/synthetic-runtime-token/);
  });
@@ -159,7 +165,11 @@ describe("memhub ui command wiring", () => {
    const output = capture();
    let code: number;
    try {
-      code = await run(["ui", "--json"], {}, { openBrowser: async () => false });
+      code = await run(
+        ["ui", "--json"],
+        {},
+        { openBrowser: async () => false },
+      );
    } finally {
      output.restore();
    }
@@ -185,7 +195,11 @@ describe("memhub ui command wiring", () => {
      const output = capture();
      let code: number;
      try {
-        code = await run(["ui", "--json"], {}, { openBrowser: async () => true });
+        code = await run(
+          ["ui", "--json"],
+          {},
+          { openBrowser: async () => true },
+        );
      } finally {
        output.restore();
      }
@@ -9,7 +9,8 @@ import { parseUiSessionExchange } from "./ui-session.js";
 describe("ui session exchange over HTTP", () => {
  let server: Server;
  let port: number;
-  const seen: Array<{ authorization?: string; method?: string; url?: string }> = [];
+  const seen: Array<{ authorization?: string; method?: string; url?: string }> =
+    [];
  let status = 200;

  beforeAll(async () => {
@@ -45,11 +45,31 @@ describe("memhub ui session wiring", () => {
    for (const doc of [
      null,
      {},
-      { schema_version: 2, exchange_code: "x", expires_in_seconds: 60, url: "http://127.0.0.1:1/ui/login?code=x" },
+      {
+        schema_version: 2,
+        exchange_code: "x",
+        expires_in_seconds: 60,
+        url: "http://127.0.0.1:1/ui/login?code=x",
+      },
      { schema_version: 1, exchange_code: "x", expires_in_seconds: 60 },
-      { schema_version: 1, exchange_code: "", expires_in_seconds: 60, url: "http://127.0.0.1:1/ui/login?code=x" },
-      { schema_version: 1, exchange_code: "x", expires_in_seconds: 0, url: "http://127.0.0.1:1/ui/login?code=x" },
-      { schema_version: 1, exchange_code: "x", expires_in_seconds: 60, url: "http://127.0.0.1:1/" },
+      {
+        schema_version: 1,
+        exchange_code: "",
+        expires_in_seconds: 60,
+        url: "http://127.0.0.1:1/ui/login?code=x",
+      },
+      {
+        schema_version: 1,
+        exchange_code: "x",
+        expires_in_seconds: 0,
+        url: "http://127.0.0.1:1/ui/login?code=x",
+      },
+      {
+        schema_version: 1,
+        exchange_code: "x",
+        expires_in_seconds: 60,
+        url: "http://127.0.0.1:1/",
+      },
    ]) {
      expect(parseUiSessionExchange(doc)).toBeNull();
    }
@@ -59,9 +79,18 @@ describe("memhub ui session wiring", () => {
    expect(uiSessionFailure("ui_web_assets_unavailable")).toMatchObject({
      code: "ui_web_assets_unavailable",
    });
-    expect(uiExitCode({ state: "opened", url: "http://127.0.0.1:1/ui/login?code=x" })).toBe(0);
-    expect(uiExitCode({ state: "blocked", code: "ui_session_contract_unavailable" })).toBe(4);
-    expect(uiExitCode({ state: "failed", code: "ui_session_code_invalid" })).toBe(1);
+    expect(
+      uiExitCode({
+        state: "opened",
+        url: "http://127.0.0.1:1/ui/login?code=x",
+      }),
+    ).toBe(0);
+    expect(
+      uiExitCode({ state: "blocked", code: "ui_session_contract_unavailable" }),
+    ).toBe(4);
+    expect(
+      uiExitCode({ state: "failed", code: "ui_session_code_invalid" }),
+    ).toBe(1);
  });

  it("uses a platform-appropriate opener and reports missing openers", async () => {
@@ -42,7 +42,12 @@ function isLoopbackHttpUrl(value: string): boolean {
  }
  if (parsed.protocol !== "http:") return false;
  const host = parsed.hostname.toLowerCase();
-  return host === "localhost" || host === "127.0.0.1" || host === "[::1]" || host === "::1";
+  return (
+    host === "localhost" ||
+    host === "127.0.0.1" ||
+    host === "[::1]" ||
+    host === "::1"
+  );
 }

 /** Validate the Core exchange document; return null when it does not match the
@@ -80,7 +80,13 @@ describe("calls clear", () => {
  });

  it("previews without contacting Core", async () => {
-    const result = await capturedRun(["calls", "clear", "--event", EVENT, "--json"]);
+    const result = await capturedRun([
+      "calls",
+      "clear",
+      "--event",
+      EVENT,
+      "--json",
+    ]);
    expect(result.code).toBe(0);
    const document = JSON.parse(result.stdout);
    expect(document.state).toBe("preview");
@@ -341,7 +347,10 @@ describe("notes / extraction / graph / manual extract", () => {
      "--json",
    ]);
    expect(applied.code).toBe(0);
-    const body = transport.post.mock.calls.at(-1)?.[1] as Record<string, unknown>;
+    const body = transport.post.mock.calls.at(-1)?.[1] as Record<
+      string,
+      unknown
+    >;
    expect(body.text).toBe("记住这个决定");
    expect(body.scope).toBe("all-apps");
    expect(JSON.parse(applied.stdout).state).toBe("committed");
@@ -10,18 +10,17 @@

 import { createHash, randomUUID } from "node:crypto";
 import {
-  MemoryNoteInputError,
  buildMemoryNoteSaveRequest,
-  parseMemoryNoteReceipt,
+  MemoryNoteInputError,
  parseExtractionSettingsReceipt,
  parseGraphWidthDecisionReceipt,
  parseManualExtractReceipt,
+  parseMemoryNoteReceipt,
 } from "@one-memory/operations";
 import type { CoreOperations, CorePutOperations } from "./core-operations.js";
 import { PRODUCT } from "./version.js";

-const UUID =
-  /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
+const UUID = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
 const DIGEST = /^[0-9a-f]{64}$/;
 const NOTE_MAX_CHARACTERS = 12_000;
 const EXTRACTION_PRESETS = ["3m", "6m", "1y", "all"] as const;
@@ -117,7 +116,10 @@ function digestOf(value: unknown): string {
 type Draft =
  | Omit<Extract<ExtendedWritePlan, { kind: "memory_note_add" }>, "digest">
  | Omit<Extract<ExtendedWritePlan, { kind: "memory_note_process" }>, "digest">
-  | Omit<Extract<ExtendedWritePlan, { kind: "extraction_window_set" }>, "digest">
+  | Omit<
+      Extract<ExtendedWritePlan, { kind: "extraction_window_set" }>,
+      "digest"
+    >
  | Omit<Extract<ExtendedWritePlan, { kind: "graph_width_decide" }>, "digest">
  | Omit<Extract<ExtendedWritePlan, { kind: "manual_extract" }>, "digest">;

@@ -148,7 +150,10 @@ function committed(
  };
 }

-function rejected(operationId: string | undefined, code: string): ExtendedWriteOutcome {
+function rejected(
+  operationId: string | undefined,
+  code: string,
+): ExtendedWriteOutcome {
  return {
    schema_version: 1,
    product: PRODUCT,
@@ -158,7 +163,10 @@ function rejected(operationId: string | undefined, code: string): ExtendedWriteO
  };
 }

-function unknown(operationId: string | undefined, code: string): ExtendedWriteOutcome {
+function unknown(
+  operationId: string | undefined,
+  code: string,
+): ExtendedWriteOutcome {
  return {
    schema_version: 1,
    product: PRODUCT,
@@ -197,7 +205,9 @@ export function validNoteText(text: string): boolean {
  );
 }

-export function planMemoryNoteProcess(evidenceId: string): ExtendedWriteOutcome {
+export function planMemoryNoteProcess(
+  evidenceId: string,
+): ExtendedWriteOutcome {
  return UUID.test(evidenceId)
    ? preview({
        schema_version: 1,
@@ -313,8 +323,7 @@ export function planGraphWidthDecide(decision: string): ExtendedWriteOutcome {

 export function planManualExtract(evidenceIds: string[]): ExtendedWriteOutcome {
  const ids = [...new Set(evidenceIds)];
-  return ids.length === evidenceIds.length &&
-    ids.every((id) => UUID.test(id))
+  return ids.length === evidenceIds.length && ids.every((id) => UUID.test(id))
    ? preview({
        schema_version: 1,
        kind: "manual_extract",
@@ -455,12 +464,12 @@ export function validateExtendedWritePlan(
 }

 export function extendedWriteExitCode(document: ExtendedWriteOutcome): number {
-  return ["preview", "committed", "recovered"].includes(document.state)
-    ? 0
-    : 1;
+  return ["preview", "committed", "recovered"].includes(document.state) ? 0 : 1;
 }

-export function renderExtendedWriteText(document: ExtendedWriteOutcome): string {
+export function renderExtendedWriteText(
+  document: ExtendedWriteOutcome,
+): string {
  if (document.state === "preview") {
    const lines = ["以下操作尚未执行；加 --confirm 后才会写入："];
    const plan = document.plan;
@@ -61,10 +61,9 @@ describe("UI write execution", () => {
    }));
    const plan = planOf(planCallEventsClear([EVENT]));
    if (plan.kind !== "call_events_clear") throw new Error("kind");
-    const result = await executeCallEventsClear(
-      plan,
-      { post } as unknown as CoreOperations,
-    );
+    const result = await executeCallEventsClear(plan, {
+      post,
+    } as unknown as CoreOperations);
    expect(result.state).toBe("committed");
    expect(post).toHaveBeenCalledWith("/api/v1/call-events/clear", {
      event_ids: [EVENT],
@@ -79,10 +78,9 @@ describe("UI write execution", () => {
    }));
    const plan = planOf(planCallEventsClearAll());
    if (plan.kind !== "call_events_clear_all") throw new Error("kind");
-    await executeCallEventsClearAll(
-      plan,
-      { post } as unknown as CoreOperations,
-    );
+    await executeCallEventsClearAll(plan, {
+      post,
+    } as unknown as CoreOperations);
    expect(post).toHaveBeenCalledWith("/api/v1/call-events/clear-all", {
      user_confirmed: true,
    });
@@ -107,7 +105,7 @@ describe("UI write execution", () => {
        committed: true,
      },
    }));
-    const del = vi.fn(async <T,>() => ({
+    const del = vi.fn(async <T>() => ({
      state: "committed" as const,
      data: {} as T,
    }));
@@ -118,10 +116,9 @@ describe("UI write execution", () => {
    const commit = planOf(planImportFileCommit(UPLOAD));
    if (commit.kind !== "import_file_commit") throw new Error("kind");
    await executeImportFileCommit(commit, operations);
-    expect(post).toHaveBeenCalledWith(
-      `/api/v1/import-files/${UPLOAD}/commit`,
-      { confirmed: true },
-    );
+    expect(post).toHaveBeenCalledWith(`/api/v1/import-files/${UPLOAD}/commit`, {
+      confirmed: true,
+    });
    const discard = planOf(planImportFileDiscard(UPLOAD));
    if (discard.kind !== "import_file_discard") throw new Error("kind");
    expect((await executeImportFileDiscard(discard, operations)).state).toBe(
@@ -133,15 +130,12 @@ describe("UI write execution", () => {
  it("treats a malformed success receipt as unknown", async () => {
    const plan = planOf(planCallEventsClear([EVENT]));
    if (plan.kind !== "call_events_clear") throw new Error("kind");
-    const result = await executeCallEventsClear(
-      plan,
-      {
-        post: vi.fn(async () => ({
-          state: "committed" as const,
-          data: { wrong: true },
-        })),
-      } as unknown as CoreOperations,
-    );
+    const result = await executeCallEventsClear(plan, {
+      post: vi.fn(async () => ({
+        state: "committed" as const,
+        data: { wrong: true },
+      })),
+    } as unknown as CoreOperations);
    expect(result).toMatchObject({
      state: "unknown",
      code: "ui_write_receipt_invalid",
@@ -14,8 +14,7 @@ import type {
 } from "./core-operations.js";
 import { PRODUCT } from "./version.js";

-const UUID =
-  /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
+const UUID = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
 const INPUT_KINDS = ["raw_conversation", "memory_content"] as const;
 const MAX_UPLOAD_BYTES = 128 * 1024 * 1024;

@@ -139,7 +138,11 @@ export function planCallEventsClear(
  ) {
    return invalid("call_event_ids_invalid");
  }
-  return preview({ schema_version: 1, kind: "call_events_clear", event_ids: [...eventIds] });
+  return preview({
+    schema_version: 1,
+    kind: "call_events_clear",
+    event_ids: [...eventIds],
+  });
 }

 export function planCallEventsClearAll(): UiWriteOutcome {
@@ -224,11 +227,7 @@ export async function readImportFileUpload(
  }
  try {
    const stats = await handle.stat();
-    if (
-      !stats.isFile() ||
-      stats.nlink !== 1 ||
-      stats.size > MAX_UPLOAD_BYTES
-    ) {
+    if (!stats.isFile() || stats.nlink !== 1 || stats.size > MAX_UPLOAD_BYTES) {
      return null;
    }
    const buffer = Buffer.alloc(Number(stats.size));
@@ -273,9 +272,7 @@ function fileUploadOperationId(
  checksum: string,
 ): string {
  const hex = createHash("sha256")
-    .update(
-      JSON.stringify([filename, inputKind, checksum]),
-    )
+    .update(JSON.stringify([filename, inputKind, checksum]))
    .digest("hex");
  return [
    hex.slice(0, 8),
@@ -325,8 +322,7 @@ export function validateUiWritePlan(value: unknown): UiWritePlan | null {
        UUID.test(candidate.operation_id) &&
        typeof candidate.path === "string" &&
        typeof candidate.filename === "string" &&
-        candidate.filename ===
-          candidate.path.split(/[/\\]/).pop() &&
+        candidate.filename === candidate.path.split(/[/\\]/).pop() &&
        /^[0-9a-f]{64}$/.test(String(candidate.sha256)) &&
        INPUT_KINDS.includes(
          candidate.input_kind as (typeof INPUT_KINDS)[number],
@@ -364,9 +360,7 @@ export async function executeCallEventsClear(
    event_ids: plan.event_ids,
    user_confirmed: true,
  });
-  return outcome(result, plan.event_ids.join(","), [
-    "removed",
-  ]);
+  return outcome(result, plan.event_ids.join(","), ["removed"]);
 }

 export async function executeCallEventsClearAll(
@@ -543,7 +537,9 @@ function outcome(
    return committed(operationId, {});
  }
  const data =
-    result.data !== null && typeof result.data === "object" && !Array.isArray(result.data)
+    result.data !== null &&
+    typeof result.data === "object" &&
+    !Array.isArray(result.data)
      ? (result.data as Record<string, unknown>)
      : null;
  if (data === null || required.some((key) => !Object.hasOwn(data, key))) {
@@ -1,10 +1,11 @@
-import { defineConfig } from "tsup";
 import { createRequire } from "node:module";
+import { defineConfig } from "tsup";

 const require = createRequire(import.meta.url);
-const { resolveBuildVersion } = require("../../scripts/release-version.cjs") as {
-  resolveBuildVersion(environment?: NodeJS.ProcessEnv, root?: string): string;
-};
+const { resolveBuildVersion } =
+  require("../../scripts/release-version.cjs") as {
+    resolveBuildVersion(environment?: NodeJS.ProcessEnv, root?: string): string;
+  };
 const releaseVersion = resolveBuildVersion(process.env);

 export default defineConfig({
@@ -18,7 +19,9 @@ export default defineConfig({
    __ONE_MEMORY_BUILD_NUMBER__: JSON.stringify(
      process.env.ONE_MEMORY_BUILD_NUMBER ?? "unknown",
    ),
-    __ONE_MEMORY_COMMIT__: JSON.stringify(process.env.ONE_MEMORY_COMMIT ?? "unknown"),
+    __ONE_MEMORY_COMMIT__: JSON.stringify(
+      process.env.ONE_MEMORY_COMMIT ?? "unknown",
+    ),
    __MEMHUB_DEFAULT_RELEASE_MANIFEST_URL_TEMPLATE__: JSON.stringify(
      process.env.MEMHUB_DEFAULT_RELEASE_MANIFEST_URL_TEMPLATE ?? "",
    ),
@@ -1,8 +1,9 @@
-import { defineConfig } from "vitest/config";
 import { readFileSync } from "node:fs";
+import { defineConfig } from "vitest/config";

-const version = JSON.parse(readFileSync(new URL("package.json", import.meta.url), "utf8"))
-  .version as string;
+const version = JSON.parse(
+  readFileSync(new URL("package.json", import.meta.url), "utf8"),
+).version as string;

 export default defineConfig({
  define: {
@@ -11,7 +11,7 @@ folder_layout = layout == 'folder'
 datas = [('src/one_memory/embedding_assets', 'one_memory/embedding_assets')]
 binaries = []
 # Loaded lazily by the API; static dependency analysis cannot discover it.
-hiddenimports = ['one_memory.hy_permission_context']
+hiddenimports = []
 tmp_ret = collect_all('hy_memory')
 datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
 tmp_ret = collect_all('chromadb')
@@ -4,7 +4,7 @@ build-backend = "uv_build"

 [project]
 name = "one-memory"
-version = "1.0.0"
+version = "1.0.4"
 description = "MemHub local-first memory core"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -12,7 +12,7 @@ dependencies = [
    "fastapi>=0.141.1",
    "filelock>=3.32.4",
    # Public SDK version validated against MemHub's permission and recall contracts.
-    "hy-memory==2.0.1",
+    "hy-memory==2.0.4",
    # Retrieval's derived vector index directly consumes these locked versions.
    "chromadb==1.5.9",
    "numpy==2.5.2",
@@ -24,7 +24,7 @@ dependencies = [
    # transitive version we do not ask for is one an unrelated upgrade could
    # drop. Deliberately NOT sentence-transformers or torch: those would take
    # the installer past 2GB, and onnxruntime alone runs the exported model.
-    "onnxruntime>=1.22.0",
+    "onnxruntime==1.23.2",
    "tokenizers>=0.21.0",
    "pydantic-settings>=2.15.0",
    "uvicorn[standard]>=0.52.4",
@@ -20,7 +20,6 @@ from copy import copy
 from datetime import UTC, datetime, timedelta
 from enum import StrEnum
 from functools import partial
-from importlib import import_module
 from pathlib import Path
 from typing import Any, Literal, cast
 from uuid import UUID
@@ -101,7 +100,13 @@ from one_memory.imports.access import (
    SourceAccessError,
    root_identity_matches,
 )
-from one_memory.imports.adapters import SourceAdapter, SyncCompleteness
+from one_memory.imports.adapters import (
+    ScanFailure,
+    ScannedEvidence,
+    ScanOptions,
+    SourceAdapter,
+    SyncCompleteness,
+)
 from one_memory.imports.candidates import (
    SourceCandidateError,
    SourceCandidateSelection,
@@ -112,7 +117,11 @@ from one_memory.imports.connections import (
    ConnectionIntentReceipt,
    SourceConnectionStore,
 )
-from one_memory.imports.grants import SourceGrantReceipt, SourceGrantStore
+from one_memory.imports.grants import (
+    SourceAuthorizationState,
+    SourceGrantReceipt,
+    SourceGrantStore,
+)
 from one_memory.imports.host_packages import CapabilityState, HostPackageCapabilities
 from one_memory.imports.jobs import (
    ImportIssueRecord,
@@ -164,7 +173,6 @@ from one_memory.memory_library import (
 )
 from one_memory.memory_notes import MemoryNoteStore, memory_note_router
 from one_memory.memory_processing import (
-    MemoryBridgeProtocol,
    MemoryProcessingSupervisor,
    _conversation_chunks,
    memory_task_timeout_seconds,
@@ -178,10 +186,8 @@ from one_memory.memory_queue import (
 )
 from one_memory.memory_server.config_state import ConfigStore
 from one_memory.memory_tasks import MemoryTask, MemoryTaskStatus, MemoryTaskStore
-from one_memory.model_content_routes import model_content_router
 from one_memory.model_presets import seed_preset_profile
 from one_memory.model_probe import ModelProber, ModelProbeResult
-from one_memory.model_processing_permission_guard import ModelProcessingPermissionLease
 from one_memory.model_profiles import (
    ActiveModelProfileError,
    ModelCreationOperationConflictError,
@@ -194,6 +200,7 @@ from one_memory.model_profiles import (
    ModelProfileStore,
    ModelProfileUnavailableError,
 )
+from one_memory.model_request_audit import AuditedModelTransport, ModelRequestAuditStore
 from one_memory.model_request_budget import BudgetedModelTransport, ModelRequestBudget
 from one_memory.onboarding import OnboardingState, OnboardingStore
 from one_memory.recall import GovernedRecallTarget, RecallReceipt, RecallRequest
@@ -285,6 +292,9 @@ class SourceResponse(BaseModel):
    candidate_revision: int = 0
    candidate_change_allowed: bool = False
    candidate_change_reason: str | None = None
+    #: Why `authorized` is false, so the UI can ask for the one thing that
+    #: would fix it instead of offering every recovery at once.
+    authorization_state: SourceAuthorizationState = SourceAuthorizationState.OK


 class PreviewSourceCandidateRequest(BaseModel):
@@ -317,6 +327,22 @@ class SourceSyncState(BaseModel):
    sync_enabled: bool


+class SedimentationRequest(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    enabled: StrictBool
+
+
+class SedimentationState(BaseModel):
+    model_config = ConfigDict(frozen=True, extra="forbid")
+
+    enabled: bool
+    #: What the switch means for the model bill and for what still runs. Read
+    #: by the app so the setting can say so without hardcoding the wording.
+    pending_retained: bool = True
+    imports_unaffected: bool = True
+
+
 class DetectionResponse(BaseModel):
    source_id: str
    detected: bool
@@ -343,7 +369,7 @@ class ConnectSourcesRequest(BaseModel):
    import_upload_id: UUID | None = None


-ONBOARDING_LARGE_HISTORY_SESSIONS = 200
+ONBOARDING_LARGE_HISTORY_USER_TURNS = 200
 ONBOARDING_HISTORY_WINDOW = timedelta(days=3)


@@ -818,6 +844,13 @@ def create_app(
                evidence=app.state.evidence_store,
            )
            app.state.memory_write_queue = memory_write_queue
+            # Builds through 1.0.2 parked pre-invocation jobs behind the
+            # retired model-content consent bit. The embedded SDK is trusted
+            # now, so make those durable jobs claimable on the first startup.
+            await asyncio.to_thread(
+                memory_write_queue.readmit_parked,
+                states=("waiting_for_permission",),
+            )
            repaired_identity_failures = await asyncio.to_thread(
                memory_write_queue.repair_pre_send_identity_failures
            )
@@ -883,6 +916,7 @@ def create_app(
            )
            app.state.import_job_store = ImportJobStore(database_path)
            app.state.onboarding_store = OnboardingStore(database_path)
+            app.state.history_import_gate_released = False
            app.state.extraction_settings_store = ExtractionSettingsStore(database_path)
            app.state.source_grant_store = SourceGrantStore(
                database_path,
@@ -905,6 +939,7 @@ def create_app(
            app.state.source_permission_locks = {
                source_id: asyncio.Lock() for source_id in app.state.source_catalog
            }
+            app.state.sedimentation_lock = asyncio.Lock()
            _follow_workbuddy_home(app, runtime_settings.source_home)
            app.state.source_detection = {}
            app.state.secret_store = secret_store or KeyringSecretStore()
@@ -960,7 +995,7 @@ def create_app(
                # Not fatal: without the tables every read checks exactly, as
                # it does today.
                app.state.memory_visibility_error = str(error)
-            app.state.model_request_budget = (
+            model_request_budget = (
                ModelRequestBudget(
                    database_path,
                    request_limit=runtime_settings.model_request_limit,
@@ -970,6 +1005,8 @@ def create_app(
                if runtime_settings.model_request_limit is not None
                else None
            )
+            app.state.model_request_budget = model_request_budget
+            app.state.model_request_audit_store = ModelRequestAuditStore(database_path)
            app.state.memory_answerer = MemoryAnswerer(
                governance=app.state.memory_governance_store,
                filter_memories=app.state.source_read_access.memories,
@@ -1008,6 +1045,23 @@ def create_app(
                governance=app.state.memory_governance_store,
                data_dir=runtime_settings.data_dir,
                extraction_settings=app.state.extraction_settings_store,
+                bridge_factory=lambda profile, api_key, data_dir: HyMemoryBridge(
+                    profile=profile,
+                    api_key=api_key,
+                    data_dir=data_dir,
+                    http_transport_factory=lambda: AuditedModelTransport(
+                        app.state.model_request_audit_store,
+                        profile_id=profile.profile_id,
+                        profile_revision=profile.revision,
+                        provider_kind=str(profile.provider_kind),
+                        model=profile.model,
+                        inner=(
+                            BudgetedModelTransport(model_request_budget)
+                            if model_request_budget is not None
+                            else None
+                        ),
+                    ),
+                ),
                # `None` means "whatever the active profile allows". Pinned to
                # 1 in `a59f4b5` as temporary load-shedding for an acceptance
                # package, and left pinned after it: a hardcoded 1 is rule 9's
@@ -1016,12 +1070,9 @@ def create_app(
                # profile concurrency 4, recall latency is unchanged while
                # precipitation goes 4x.
                max_concurrent_tasks=runtime_settings.precipitation_max_concurrent_tasks,
-                model_content_config=app.state.config_store,
-                permission_bridge_factory=lambda lease, secret, directory: _permission_context(
-                    app, lease, secret, directory
-                ),
                source_writes_permitted=lambda source_id: _source_writes_on(app, source_id),
                service_levels=app.state.service_levels,
+                sedimentation_enabled=lambda: _config_store(app).sedimentation_enabled(),
            )
            # C2 owns the single production budget (handle 5s + reserve 250ms,
            # total 5.25s, lease 10s, renew 1s, claim 250ms). The composition
@@ -1039,18 +1090,27 @@ def create_app(
                        job.job_id,
                    )
                await asyncio.to_thread(
-                    memory_write_queue.retry_timed_out_unknown_after_reconcile,
-                    older_than=datetime.now(UTC) - timedelta(minutes=5),
+                    memory_write_queue.retry_first_unknown_after_reconcile,
+                    older_than=datetime.now(UTC) - timedelta(seconds=10),
                    limit=32,
                )
+                await asyncio.to_thread(
+                    app.state.memory_task_store.retry_first_unknown_after_reconcile,
+                    older_than=datetime.now(UTC) - timedelta(seconds=10),
+                    limit=128,
+                )

            async def admit_hy_invocation(
                source_id: str,
                mark_invoked: Callable[[], Awaitable[bool]],
            ) -> bool:
                """Serialize pause with the last boundary before an HY call."""
-                lock = app.state.source_permission_locks.setdefault(source_id, asyncio.Lock())
-                async with lock:
+                source_lock = app.state.source_permission_locks.setdefault(
+                    source_id, asyncio.Lock()
+                )
+                async with app.state.sedimentation_lock, source_lock:
+                    if not _config_store(app).sedimentation_enabled():
+                        raise RuntimeError("model_sedimentation_disabled")
                    if not _source_writes_on(app, source_id):
                        raise RuntimeError("model_content_source_disabled")
                    return await mark_invoked()
@@ -1088,8 +1148,18 @@ def create_app(
                    + 10
                ),
                reconcile_unknown=reconcile_unknown_writes,
-                source_writes_permitted=lambda source_id: _source_writes_on(app, source_id),
-                source_claim_enabled=lambda source_id: _config_store(app).sync_enabled(source_id),
+                source_writes_permitted=lambda source_id: _sedimentation_claim_enabled(
+                    app, source_id
+                ),
+                # The claim gate has to be the same question the write gate
+                # asks, not a second answer to it. It used to be
+                # `sync_enabled`, the per-app toggle, and `one-memory-manual` is
+                # not a connected app: it has no entry in the configuration
+                # document, so that predicate answers False for it while the
+                # write gate admits it by its grant. Every manual note's job was
+                # therefore skipped on every drain tick — the note stayed
+                # "saved, waiting" with no task behind it, forever.
+                source_claim_enabled=lambda source_id: _sedimentation_claim_enabled(app, source_id),
            )
            app.state.memory_write_drain = memory_write_drain
            app.state.memory_write_drain_task = watched_task(
@@ -1118,10 +1188,7 @@ def create_app(
                for chunk in _conversation_chunks(conversation, max_turns=12, completed_only=True):
                    if (
                        record not in chunk
-                        or not any(
-                            item.role == "user" and item.written_by.value == "person"
-                            for item in chunk
-                        )
+                        or not any(item.role == "user" for item in chunk)
                        or not any(item.role == "assistant" for item in chunk)
                    ):
                        continue
@@ -1134,7 +1201,6 @@ def create_app(
            app.state.matter_discovery.processor = app.state.memory_processing_supervisor
            app.state.matter_discovery.write_intake = app.state.memory_write_intake
            app.state.matter_discovery.write_queue = memory_write_queue
-            app.state.processing_coordinator = None
            app.state.memory_note_store = MemoryNoteStore(
                app.state.evidence_store,
                app.state.memory_processing_supervisor,
@@ -1446,7 +1512,6 @@ def create_app(
    app.include_router(matter_router(protected))
    app.include_router(memory_note_router(protected))
    app.include_router(file_import_router(protected))
-    app.include_router(model_content_router(protected))
    app.include_router(context_router(protected))

    def require_workbuddy_boundary(request: Request) -> None:
@@ -1736,6 +1801,30 @@ def create_app(
            "events": events,
        }

+    @app.get("/api/v1/diagnostics/model-usage", dependencies=protected)
+    def diagnostic_model_usage(
+        since: datetime | None = Query(default=None),  # noqa: B008
+        request_id: str | None = Query(  # noqa: B008
+            default=None, min_length=1, max_length=512
+        ),
+        limit: int = Query(default=200, ge=1, le=2_000),  # noqa: B008
+    ) -> dict[str, object]:
+        """Return content-free precipitation usage for provider reconciliation."""
+
+        normalized_since = (
+            (since if since.tzinfo is not None else since.replace(tzinfo=UTC))
+            if since is not None
+            else None
+        )
+        report = app.state.model_request_audit_store.report(
+            since=normalized_since.astimezone(UTC).isoformat()
+            if normalized_since is not None
+            else None,
+            request_id=request_id,
+            limit=limit,
+        )
+        return {"schema_version": 1, **report}
+
    @app.get(
        "/api/v1/runtime/identity",
        response_model=HealthResponse,
@@ -1873,6 +1962,20 @@ def create_app(
        """Record UI completion only; finishing or skipping is never model consent."""
        return await asyncio.to_thread(_onboarding_store(app).complete)

+    @app.post(
+        "/api/v1/onboarding/reset",
+        response_model=OnboardingState,
+        dependencies=protected,
+    )
+    async def reset_onboarding() -> OnboardingState:
+        """Go back to the first-run screens. Only the completion flag moves.
+
+        Memories, sources and settings live in other tables and are untouched,
+        which is the point: someone who wants to see the cold-start flow again
+        should not have to delete anything to do it.
+        """
+        return await asyncio.to_thread(_onboarding_store(app).reset)
+
    @app.get(
        "/api/v1/sources",
        response_model=list[SourceResponse],
@@ -1910,6 +2013,11 @@ def create_app(
                    candidate_change_reason=app.state.source_candidate_store.change_reason(
                        registration
                    ),
+                    authorization_state=grant_store.authorization_state(
+                        user_id="local",
+                        source_id=source_id,
+                        candidate=registration.candidate.path,
+                    ),
                )
                for source_id, registration in _catalog(app).items()
            ]
@@ -2221,24 +2329,39 @@ def create_app(
                raise HTTPException(status_code=409, detail=error.code) from None
            if grant is None:
                raise HTTPException(status_code=409, detail="grant_required")
-            await _prepare_source_resume(source_id)
+            try:
+                await _prepare_source_resume(source_id)
+            except BaseException as error:
+                raise _sync_refused("sync_prepare_failed") from error
        else:
-            await _live_sync(app).stop_source(source_id)
+            try:
+                await _live_sync(app).stop_source(source_id)
+            except BaseException as error:
+                raise _sync_refused("sync_prepare_failed") from error
        try:
            await asyncio.to_thread(_config_store(app).set_sync, source_id, request.enabled)
-        except BaseException:
+        except BaseException as error:
            if request.enabled:
                await _live_sync(app).stop_source(source_id)
            else:
                await _prepare_source_resume(source_id)
                await _activate_source_resume(source_id)
            await _refresh_live_watch(app)
-            raise
+            raise _sync_refused("sync_config_write_failed") from error
        try:
            if request.enabled:
-                await _activate_source_resume(source_id)
-            await _refresh_live_watch(app)
+                try:
+                    await _activate_source_resume(source_id)
+                except BaseException as error:
+                    raise _sync_refused("sync_resume_failed") from error
+            try:
+                await _refresh_live_watch(app)
+            except BaseException as error:
+                raise _sync_refused("sync_watcher_failed") from error
        except BaseException:
+            # Rolls back the desired state and settles the watcher exactly as
+            # before; the only change is that the raise above now carries a
+            # code saying which stage failed.
            await asyncio.to_thread(_config_store(app).set_sync, source_id, previous)
            if request.enabled:
                await _live_sync(app).stop_source(source_id)
@@ -2328,7 +2451,7 @@ def create_app(
        await asyncio.to_thread(
            app.state.memory_task_store.park_source_for_permission,
            source_id=source_id,
-            error_code="model_content_permission_required",
+            error_code="hy_memory_source_revoked",
        )
        await _live_sync(app).stop_source(source_id)
        _supervisor(app).unregister_adapter(source_id)
@@ -2369,6 +2492,8 @@ def create_app(
    async def connect_sources(
        request: ConnectSourcesRequest,
    ) -> ConnectSourcesResponse:
+        if request.onboarding:
+            app.state.history_import_gate_released = False
        source_ids = tuple(dict.fromkeys(request.source_ids))
        registrations = [_require_registration(app, source_id) for source_id in source_ids]
        if request.request_id is None:
@@ -3534,6 +3659,48 @@ def create_app(
    def put_extraction_settings(request: ExtractionWindowUpdate) -> ExtractionSettings:
        return _memory_processing_supervisor(app).set_extraction_window(request.window_preset)

+    @app.get(
+        "/api/v1/sedimentation",
+        response_model=SedimentationState,
+        dependencies=protected,
+    )
+    def get_sedimentation() -> SedimentationState:
+        """Whether the model may still turn conversations into memories."""
+        return SedimentationState(enabled=_config_store(app).sedimentation_enabled())
+
+    @app.put(
+        "/api/v1/sedimentation",
+        response_model=SedimentationState,
+        dependencies=protected,
+    )
+    async def put_sedimentation(request: SedimentationRequest) -> SedimentationState:
+        """Start or stop model-driven sedimentation, keeping the configuration.
+
+        Stopping is the user's cost control: nothing further goes to the model
+        and already queued work waits rather than failing, so starting again
+        resumes it. Importing conversations is not affected either way.
+        """
+        store = _config_store(app)
+        changed = False
+        async with app.state.sedimentation_lock:
+            previous = store.sedimentation_enabled()
+            if previous is not request.enabled:
+                await asyncio.to_thread(store.set_sedimentation_enabled, request.enabled)
+                changed = True
+                diagnostic_event(
+                    level="info",
+                    component="memory.sedimentation",
+                    event="memory_sedimentation.toggled",
+                    outcome="succeeded",
+                    details={"enabled": request.enabled},
+                )
+        if changed and request.enabled:
+            # Work parked by the stop is picked up again rather than waiting
+            # for the next natural dispatch, but not on this request's time:
+            # see _resume_sedimentation.
+            _resume_sedimentation(app)
+        return SedimentationState(enabled=store.sedimentation_enabled())
+
    @app.post(
        "/api/v1/memory-processing/manual-extract",
        response_model=ManualExtractReceipt,
@@ -3894,12 +4061,8 @@ async def _authorize_answer_generation(
    # The explicit question is this operation's input, not a standing permission for history.
    if not question.strip() or not grounding:
        raise MemoryAnswerError("memory_answer_generation_not_authorized")
-    if not await asyncio.to_thread(_config_store(app).model_content_allowed):
-        raise MemoryAnswerError("memory_answer_generation_not_authorized")

    def check() -> None:
-        if not _config_store(app).model_content_allowed():
-            raise MemoryAnswerError("memory_answer_generation_not_authorized")
        if _model_profile_store(app).active_identity() != (
            profile.profile_id,
            profile.revision,
@@ -3923,29 +4086,6 @@ async def _authorize_answer_generation(
    return recheck


-@asynccontextmanager
-async def _permission_context(
-    app: FastAPI,
-    lease: ModelProcessingPermissionLease,
-    secret: str | None,
-    directory: Path,
-) -> AsyncIterator[MemoryBridgeProtocol]:
-    module = await asyncio.to_thread(import_module, "one_memory.hy_permission_context")
-    budget = cast(ModelRequestBudget | None, app.state.model_request_budget)
-    source_id = lease.records[0].source_id if lease.records else "manual"
-    factory = module.permission_bridge_factory(
-        lambda profile, api_key, data_dir: _memory_processing_supervisor(app)._bridge(
-            profile,
-            api_key,
-            data_dir,
-            scope_key=source_id,
-        ),
-        **({"http_transport": BudgetedModelTransport(budget)} if budget else {}),
-    )
-    async with factory(lease, secret, directory) as bridge:
-        yield cast(MemoryBridgeProtocol, bridge)
-
-
 class _LazyASGIApp:
    """Compatibility ASGI app that is built only when a server imports it.

@@ -3977,6 +4117,16 @@ class _LazyASGIApp:
 app = _LazyASGIApp()


+def _sync_refused(code: str) -> HTTPException:
+    """A sync refusal that says which stage refused.
+
+    Turning the switch on takes four steps: authorize, prepare, write, resume.
+    The UI used to be told only that the switch had sprung back, and every
+    stage needs a different fix, so each one gets its own code.
+    """
+    return HTTPException(status_code=409, detail=code)
+
+
 def _granted_roots(
    source_ids: Iterable[str],
    grant_store: SourceGrantStore,
@@ -4028,6 +4178,7 @@ def _source_response(
    totals: tuple[int, int],
    *,
    candidate_change_reason: str | None = None,
+    authorization_state: SourceAuthorizationState = SourceAuthorizationState.OK,
 ) -> SourceResponse:
    location = inspect_source_location(
        registration.descriptor.source_id, registration.candidate.path
@@ -4240,7 +4391,7 @@ async def _onboarding_history_after(
    home: Path,
    now: datetime | None = None,
 ) -> datetime | None:
-    """Return one shared cutoff when any selected harness has large history.
+    """Return one shared cutoff when any selected harness has large, old history.

    The cutoff is decided before any history job is created, then persisted on
    every accepted job. This is deliberately global: independently limiting
@@ -4249,6 +4400,8 @@ async def _onboarding_history_after(
    """
    if not request.onboarding:
        return None
+    anchor = now or datetime.now(UTC)
+    cutoff = anchor - ONBOARDING_HISTORY_WINDOW
    revisions = request.source_candidate_revisions or {}
    for registration in registrations:
        source_id = registration.descriptor.source_id
@@ -4276,29 +4429,52 @@ async def _onboarding_history_after(
                home=home,
            )
            adapter = registration.create_adapter(grant)
-            count_sessions = cast(
-                Callable[..., Awaitable[int]] | None,
-                getattr(adapter, "count_sessions", None),
-            )
-            if count_sessions is not None:
-                session_count = await count_sessions(limit=ONBOARDING_LARGE_HISTORY_SESSIONS + 1)
-            else:
-                pending = _install_state(adapter)
-                if pending is None:
-                    continue
-                session_count = (await pending).session_count
+            if await _has_large_old_history(adapter, cutoff=cutoff):
+                return cutoff
        except (OSError, SourceAccessError):
            # A supported source whose metadata count cannot be completed is
            # treated conservatively. The later connection still owns the
            # precise authorization error, while onboarding never falls back
            # to an unbounded history import.
-            session_count = ONBOARDING_LARGE_HISTORY_SESSIONS + 1
-        if session_count > ONBOARDING_LARGE_HISTORY_SESSIONS:
-            anchor = now or datetime.now(UTC)
-            return anchor - ONBOARDING_HISTORY_WINDOW
+            return cutoff
    return None


+async def _has_large_old_history(adapter: SourceAdapter, *, cutoff: datetime) -> bool:
+    """Probe one authorized Harness with the same parsers the import will use.
+
+    Stop as soon as both policy facts are proven: more than 200 genuine person
+    turns and at least one record older than the three-day cutoff. Counting
+    parsed Evidence instead of transcript files prevents one giant coding
+    session from bypassing the limit, while reusing the adapters keeps host
+    envelopes and tool-shaped user records out of the count.
+    """
+
+    user_turns = 0
+    has_old_data = False
+    async for event in adapter.scan(ScanOptions()):
+        if isinstance(event, ScanFailure):
+            # A partial probe must never turn into an unbounded first import.
+            return True
+        if not isinstance(event, ScannedEvidence):
+            continue
+        observed = event.evidence.created_at
+        if observed is None and event.cursor.observed_modified_ns > 0:
+            observed = datetime.fromtimestamp(
+                event.cursor.observed_modified_ns / 1_000_000_000,
+                tz=UTC,
+            )
+        if observed is not None:
+            if observed.tzinfo is None:
+                observed = observed.replace(tzinfo=UTC)
+            has_old_data = has_old_data or observed.astimezone(UTC) < cutoff
+        if event.evidence.role == "user":
+            user_turns += 1
+        if user_turns > ONBOARDING_LARGE_HISTORY_USER_TURNS and has_old_data:
+            return True
+    return False
+
+
 def _install_state(adapter: object) -> Awaitable[SourceInstallState] | None:
    """Ask an adapter for the two-level answer when it has one.

@@ -4590,7 +4766,7 @@ async def _keep_loop_heartbeat(watchdog: LoopWatchdog) -> None:
 async def _emit_resource_snapshots(interval_seconds: float = 60.0) -> None:
    while True:
        diagnostic_event(
-            level="info",
+            level="debug",
            component="core.resources",
            event="core.resource_snapshot",
            outcome="observed",
@@ -4615,6 +4791,33 @@ def _start_backfill(app: FastAPI) -> None:
    )


+def _resume_sedimentation(app: FastAPI) -> None:
+    """Pick the work parked by a stop back up, off the caller's clock.
+
+    Turning the switch back on is a decision, not a job the caller waits for:
+    the choice is already stored, and what follows it is maintenance over every
+    unfinished task followed by a scan of the whole queue. On a real backlog
+    that runs for many minutes, so awaiting it here left the PUT unanswered
+    until the client gave up — the renderer showed "启用未能确认" for a switch
+    that had in fact been saved, and the switch only appeared to turn on once
+    the follow-up read came back. Measured: one toggle, twenty-plus minutes,
+    no response, client timeout at thirty seconds.
+
+    Repeated toggles reuse the running walk; it reads the switch as it goes, so
+    a stop part-way through is honoured on the next batch instead of needing a
+    cancel.
+    """
+    running = cast("asyncio.Task[None] | None", getattr(app.state, "sedimentation_resume", None))
+    if running is not None and not running.done():
+        return
+    app.state.sedimentation_resume = watched_task(
+        asyncio.get_running_loop(),
+        lambda: _memory_processing_supervisor(app).resume_allowed(),
+        component="memory.sedimentation",
+        name="one-memory-sedimentation-resume",
+    )
+
+
 def _start_visibility_recompute(app: FastAPI) -> None:
    """Keep the stored visibility conclusions current, in the background.

@@ -4723,8 +4926,6 @@ async def _run_backfill(app: FastAPI) -> None:
    app.state.memory_backfill_error = None
    if not app.state.settings.automatic_memory_processing:
        return
-    if not await asyncio.to_thread(_config_store(app).model_content_allowed):
-        return
    try:
        queue = app.state.memory_write_queue
        intake = app.state.memory_write_intake
@@ -4860,6 +5061,51 @@ def _source_writes_on(app: FastAPI, source_id: str) -> bool:
    return _source_processing_enabled(app, source_id)


+def _sedimentation_claim_enabled(app: FastAPI, source_id: str) -> bool:
+    """Claim queued model work only while both global and source gates allow it."""
+
+    return (
+        _config_store(app).sedimentation_enabled()
+        and _history_import_gate_released(app)
+        and _source_writes_on(app, source_id)
+    )
+
+
+def _history_import_gate_released(app: FastAPI) -> bool:
+    """Hold first sedimentation until every accepted history import is terminal."""
+
+    if getattr(app.state, "history_import_gate_released", False):
+        return True
+    onboarding = getattr(app.state, "onboarding_store", None)
+    jobs = getattr(app.state, "import_job_store", None)
+    if onboarding is None or jobs is None:
+        return True
+    state = onboarding.get()
+    if state.status != "importing" or not state.selected_source_ids:
+        app.state.history_import_gate_released = True
+        return True
+    selected = set(state.selected_source_ids)
+    matching = [
+        job
+        for job in jobs.list_jobs()
+        if job.operation == "history"
+        and job.source_id in selected
+        and (state.started_at is None or job.created_at >= state.started_at)
+    ]
+    if {job.source_id for job in matching} != selected:
+        return False
+    active = {
+        ImportStatus.PENDING,
+        ImportStatus.RUNNING,
+        ImportStatus.PAUSED,
+        ImportStatus.INTERRUPTED,
+    }
+    released = all(job.status not in active for job in matching)
+    if released:
+        app.state.history_import_gate_released = True
+    return released
+
+
 def _model_prober(app: FastAPI) -> ModelProber:
    return cast(ModelProber, app.state.model_prober)

@@ -175,6 +175,22 @@ async def in_interactive_read[T](
    return await in_read_pool(interactive_read_pool(), work, *args, **kwargs)


+def _retrieve[T](future: "asyncio.Future[T]") -> None:
+    """Mark a shielded read's outcome as seen.
+
+    `in_read_pool` shields the wrapped worker, so a caller that gives up and
+    cancels leaves the read running to completion. When that read then fails,
+    nothing awaits the future any more: asyncio reports the orphan at GC as
+    "Future exception was never retrieved", the loop handler turns it into a
+    `core.loop.task_exception` error diagnostic, and a lane that degraded on
+    its own budget -- correctly, and already reported -- looks like a crashed
+    task on top of it. Retrieving here keeps the diagnostics honest. Awaiting
+    still raises for callers that are still there, so no failure is hidden.
+    """
+    if not future.cancelled():
+        future.exception()
+
+
 async def in_read_pool[T](
    pool: ThreadPoolExecutor,
    work: Callable[..., T],
@@ -193,7 +209,9 @@ async def in_read_pool[T](
            _interactive_reads -= 1

    worker.add_done_callback(released)
-    return await asyncio.shield(asyncio.wrap_future(worker))
+    pending = asyncio.wrap_future(worker)
+    pending.add_done_callback(_retrieve)
+    return await asyncio.shield(pending)


 async def wait_for_interactive_reads() -> None:
@@ -266,7 +266,13 @@ def serve(*, maintenance: bool = False, reader: bool = False) -> None:
        )
        application = create_reader_app(reader_settings)
        print(f"reader_port:{port}", file=sys.stderr, flush=True)
-        uvicorn.run(application, host=settings.host, port=port, reload=False)
+        uvicorn.run(
+            application,
+            host=settings.host,
+            port=port,
+            reload=False,
+            access_log=False,
+        )
        return

    from one_memory.api import create_app
@@ -297,9 +303,21 @@ def serve(*, maintenance: bool = False, reader: bool = False) -> None:
    )
    if maintenance:
        with CoreRuntime(settings, publish_receipt=False):
-            uvicorn.run(application, host=settings.host, port=settings.port, reload=False)
+            uvicorn.run(
+                application,
+                host=settings.host,
+                port=settings.port,
+                reload=False,
+                access_log=False,
+            )
    else:
-        uvicorn.run(application, host=settings.host, port=settings.port, reload=False)
+        uvicorn.run(
+            application,
+            host=settings.host,
+            port=settings.port,
+            reload=False,
+            access_log=False,
+        )


 if __name__ == "__main__":
@@ -196,7 +196,7 @@ def read_raw_context(
        if _prefers_declared_facts(request.query)
        else "1"
    )
-    person_authored = "CASE WHEN e.written_by='person' THEN 1 ELSE 0 END"
+    person_authored = "CASE WHEN e.role='user' THEN 1 ELSE 0 END"
    query_echo = (
        "CASE WHEN trim(e.verbatim)=? OR trim(e.verbatim)=? THEN 0 ELSE 1 END"
        if _prefers_declared_facts(request.query)
@@ -470,9 +470,8 @@ def revalidate_raw_context(
                    )
                except (ValueError, TypeError):
                    break
-                if (row["role"], row["written_by"], created) != (
+                if (row["role"], created) != (
                    message.role,
-                    message.written_by,
                    message.created_at,
                ):
                    break
@@ -66,12 +66,12 @@ def _field(text: str) -> ResumeField | None:
    return None


-def _raw_class(written_by: str, evidence_id: str, accepted: frozenset[str]) -> EvidenceClass:
-    if written_by == "person":
+def _raw_class(role: str, evidence_id: str, accepted: frozenset[str]) -> EvidenceClass:
+    if role == "user":
        return "person_statement"
-    if written_by == "agent":
+    if role == "assistant":
        return "agent_report"
-    if written_by == "tool" and evidence_id in accepted:
+    if role == "tool" and evidence_id in accepted:
        return "system_acceptance"
    return "tool_output"

@@ -118,7 +118,7 @@ def assemble_resume_packet(
                    field=_field(message.text) or "unclassified",
                    text=message.text,
                    evidence_class=_raw_class(
-                        message.written_by,
+                        message.role,
                        message.evidence_id,
                        verified_system_evidence_ids,
                    ),
@@ -345,7 +345,7 @@ def _lexical_raw_rows(
        if context_raw._prefers_declared_facts(query.query)
        else "1"
    )
-    person_authored = "CASE WHEN e.written_by='person' THEN 1 ELSE 0 END"
+    person_authored = "CASE WHEN e.role='user' THEN 1 ELSE 0 END"
    query_echo = (
        "CASE WHEN trim(e.verbatim)=? OR trim(e.verbatim)=? THEN 0 ELSE 1 END"
        if context_raw._prefers_declared_facts(query.query)
@@ -507,7 +507,7 @@ def _raw_item(row: sqlite3.Row, start1: int, matched_by: tuple[str, ...]) -> Raw
            "source_id": row["source_id"],
            "conversation_id": row["conversation_id"],
            "role": row["role"],
-            "written_by": "person" if row["written_by"] == "person" else "agent",
+            "written_by": "person" if row["role"] == "user" else "agent",
            "event_time": event_time.isoformat() if event_time else None,
            "spans": [{"start_cp": start, "end_cp": start + len(text), "text": text}],
            "matched_by": list(matched_by),
@@ -1039,7 +1039,7 @@ def close_main_databases() -> None:
 def log_main_database_stats(moment: str) -> None:
    """One line in core.log: how many connections are open and who waited."""
    stats = main_database_stats()
-    logger.info(
+    logger.debug(
        "main_database_connections moment=%s %s",
        moment,
        " ".join(f"{key}={value}" for key, value in stats.items()),
@@ -19,6 +19,7 @@ from __future__ import annotations

 import atexit
 import json
+import logging
 import os
 import re
 import sys
@@ -139,6 +140,10 @@ _DETAIL_KEYS = frozenset(
        "accounted_ms",
        "unaccounted_ms",
        "conversation_hash",
+        # The sedimentation switch: a bounded boolean, the same shape as
+        # `spawned` and `index_complete`. Left out of this list it did not
+        # drop a log line, it turned the switch's own request into a 500.
+        "enabled",
    }
 )
 _trace_id: ContextVar[str | None] = ContextVar("one_memory_trace_id", default=None)
@@ -649,12 +654,24 @@ def diagnostic_event(**values: object) -> None:
    active = _active
    if active is None:
        return
+    if values.get("level") == "debug" and not logging.getLogger().isEnabledFor(
+        logging.DEBUG
+    ):
+        return
    try:
        active.write(**values)  # type: ignore[arg-type]
    except OSError:
        # Diagnostics must not become a new product outage when a disk is full
        # or the log directory is temporarily unavailable.
        return
+    except ValueError:
+        # Same promise, same reason, for a rejected payload instead of a failed
+        # write. An unlisted `details` key once turned every external recall
+        # into a 500 and then killed the process, and the sedimentation switch
+        # into a permanent failure. The payload itself is still wrong and is
+        # still caught in CI by tests/test_diagnostic_payload_keys.py; what
+        # must not happen is the caller paying for it.
+        return


 def current_session_id() -> str | None:
@@ -21,7 +21,6 @@ from pydantic import BaseModel, ConfigDict, Field

 from one_memory.diagnostic_log import diagnostic_event, diagnostic_exception, elapsed_ms
 from one_memory.hy_lifecycle import HyLifecycle, HyLifecycleUnreadable, read_hy_lifecycle
-from one_memory.hy_permission_store import native_store_read_slot
 from one_memory.hy_presence import HyPresence, confirm_deleted
 from one_memory.hy_runtime_config import (
    HY_MEMORY_MODE,
@@ -32,7 +31,9 @@ from one_memory.hy_runtime_config import (
    diagnostic_log_level,
 )
 from one_memory.hy_source_context import HySourceContext, HyWriteBindings
+from one_memory.hy_store_concurrency import native_store_read_slot
 from one_memory.model_profiles import ModelProfileSnapshot, ProviderKind, attach_disable_thinking
+from one_memory.model_request_audit import audit_precipitation_request
 from one_memory.semantic_deadline import SemanticMemoryHit

 #: D102: how relevant HY has to call a hit before it counts. Raised from 0.25
@@ -143,6 +144,22 @@ class HyMemoryBridgeError(RuntimeError):
    pass


+MEMHUB_DEFAULT_CODING_TOOL_RESULT_MAX_BYTES = 512
+
+
+def coding_tool_result_max_bytes() -> int:
+    """Return the per-tool payload ceiling used by MemHub's HY client."""
+
+    raw = os.environ.get("MEMORY_CODING_TOOL_RESULT_MAX_BYTES")
+    if raw is None:
+        return MEMHUB_DEFAULT_CODING_TOOL_RESULT_MAX_BYTES
+    try:
+        value = int(raw)
+    except ValueError:
+        return MEMHUB_DEFAULT_CODING_TOOL_RESULT_MAX_BYTES
+    return value if value > 0 else MEMHUB_DEFAULT_CODING_TOOL_RESULT_MAX_BYTES
+
+
 def _exclusive[**P, T](
    method: Callable[P, Coroutine[Any, Any, T]],
 ) -> Callable[P, Coroutine[Any, Any, T]]:
@@ -376,11 +393,13 @@ class HyMemoryBridge:
        api_key: str | None,
        data_dir: Path,
        client: HyMemoryClientProtocol | None = None,
+        http_transport_factory: Callable[[], Any] | None = None,
    ) -> None:
        self.profile = profile
        self.api_key = api_key
        self.data_dir = data_dir
        self._client = client
+        self._http_transport_factory = http_transport_factory
        self._client_lock = asyncio.Lock()
        self._write_lock = asyncio.Lock()
        self._last_bindings: HyWriteBindings | None = None
@@ -412,8 +431,9 @@ class HyMemoryBridge:

    @_exclusive
    async def ingest(self, request: HyMemoryIngestRequest) -> HyMemoryIngestResult:
-        async with self._write_lock:
-            return await self._ingest(request)
+        with audit_precipitation_request(request.request_id):
+            async with self._write_lock:
+                return await self._ingest(request)

    async def _ingest(self, request: HyMemoryIngestRequest) -> HyMemoryIngestResult:
        started = monotonic()
@@ -427,6 +447,11 @@ class HyMemoryBridge:
        )
        client = await self._get_client()
        guards = getattr(client, "_one_memory_write_guards", ())
+        from one_memory.hy_response_guard import HyReconcileGuard
+
+        reconcile_guards = tuple(guard for guard in guards if isinstance(guard, HyReconcileGuard))
+        for guard in reconcile_guards:
+            guard.begin_write()
        before = [len(guard.failures) for guard in guards]
        bindings = HyWriteBindings(request.source_context) if request.source_context else None
        if bindings:
@@ -500,6 +525,12 @@ class HyMemoryBridge:
            if len(guard.failures) > count:
                raise HyMemoryBridgeError(guard.failures[count])
        result = await self._result(client, request, raw, expected_contents=None, bindings=bindings)
+        result = await _with_native_mutation_delta(
+            client,
+            result,
+            raw,
+            reconcile_guards,
+        )
        if bindings:
            outputs = []
            for memory in result.memories:
@@ -529,8 +560,9 @@ class HyMemoryBridge:
        request: HyMemoryIngestRequest,
        contents: Sequence[str],
    ) -> HyMemoryIngestResult:
-        async with self._write_lock:
-            return await self._ingest_extracted(request, contents)
+        with audit_precipitation_request(request.request_id):
+            async with self._write_lock:
+                return await self._ingest_extracted(request, contents)

    async def _ingest_extracted(
        self,
@@ -553,6 +585,8 @@ class HyMemoryBridge:
        from one_memory.hy_response_guard import HyReconcileGuard

        report_guards = tuple(guard for guard in guards if isinstance(guard, HyReconcileGuard))
+        for guard in report_guards:
+            guard.begin_write()
        expected_contents = frozenset(content.strip() for content in contents if content.strip())
        for guard in report_guards:
            guard.imported_contents = expected_contents
@@ -599,6 +633,12 @@ class HyMemoryBridge:
            raw,
            expected_contents=expected_contents,
        )
+        result = await _with_native_mutation_delta(
+            client,
+            result,
+            raw,
+            report_guards,
+        )
        diagnostic_event(
            level="info",
            component="hy.bridge",
@@ -1087,6 +1127,7 @@ class HyMemoryBridge:
                                self.profile,
                                api_key=self.api_key,
                                data_dir=self.data_dir,
+                                http_transport_factory=self._http_transport_factory,
                            )

                    await _joined_thread(initialize)
@@ -1113,7 +1154,7 @@ class HyMemoryBridge:
                    )
                    raise HyMemoryBridgeError("hy_memory_init_failed") from error
                diagnostic_event(
-                    level="info",
+                    level="debug",
                    component="hy.client",
                    event="hy.client_initialization.completed",
                    outcome="succeeded",
@@ -1136,6 +1177,7 @@ def create_hy_memory_client(
    *,
    api_key: str | None,
    data_dir: Path,
+    http_transport_factory: Callable[[], Any] | None = None,
 ) -> HyMemoryClientProtocol:
    configure_hy_memory_environment(data_dir)
    config = build_hy_memory_config(
@@ -1143,6 +1185,7 @@ def create_hy_memory_client(
        api_key=api_key,
        data_dir=data_dir,
    )
+    config.http_transport_factory = http_transport_factory
    reattach_legacy_vector_collection(profile, config)
    mismatch = inspect_graph_width_mismatch(config)
    if mismatch is not None:
@@ -1287,6 +1330,72 @@ async def _coding_result(
    return HyMemoryIngestResult(success=True, memories=tuple(outputs))


+async def _with_native_mutation_delta(
+    client: HyMemoryClientProtocol,
+    result: HyMemoryIngestResult,
+    raw: Mapping[str, object],
+    reconcile_guards: Sequence[Any],
+) -> HyMemoryIngestResult:
+    """Map native SDK mutations back into MemHub governance.
+
+    Store permission wrappers used to discover this by intercepting every
+    native CRUD call. The embedded SDK now owns unrestricted local CRUD, so
+    chat targets come from the reconcile observer and coding targets come from
+    the SDK's public result document.
+    """
+
+    retired = set(result.retired_memory_ids)
+    superseded = {key: tuple(values) for key, values in result.superseded_memory_ids.items()}
+    chat_targets = {
+        key
+        for guard in reconcile_guards
+        for key in getattr(guard, "mutation_target_ids", ())
+        if isinstance(key, str) and key
+    }
+
+    operations = raw.get("ops")
+    for item in operations if isinstance(operations, list) else ():
+        if not isinstance(item, Mapping):
+            continue
+        if item.get("action") == "DELETE":
+            key = item.get("target_experience_id")
+            if isinstance(key, str) and key:
+                retired.add(key)
+
+    fact_operations = raw.get("fact_ops")
+    for item in fact_operations if isinstance(fact_operations, list) else ():
+        if not isinstance(item, Mapping) or item.get("action") != "SUPERSEDE":
+            continue
+        old, new = item.get("fact_id"), item.get("superseded_by")
+        if isinstance(old, str) and old:
+            retired.add(old)
+            if isinstance(new, str) and new:
+                superseded[old] = (new,)
+
+    for key in chat_targets:
+        record = await _joined_thread(_read_hy_record, client, key)
+        if record is None:
+            retired.add(key)
+            continue
+        status = str(record.get("status", "")).lower()
+        if status in {"deleted", "shadow", "superseded"}:
+            retired.add(key)
+        replacement = record.get("superseded_by")
+        if isinstance(replacement, str) and replacement:
+            superseded[key] = (replacement,)
+        elif isinstance(replacement, list):
+            values = tuple(value for value in replacement if isinstance(value, str) and value)
+            if values:
+                superseded[key] = values
+
+    return result.model_copy(
+        update={
+            "retired_memory_ids": tuple(sorted(retired)),
+            "superseded_memory_ids": superseded,
+        }
+    )
+
+
 def _read_hy_record(
    client: HyMemoryClientProtocol,
    memory_id: str,
@@ -1603,6 +1712,7 @@ def build_hy_memory_config(
    config.history.db_path = str(memory_root / "history.sqlite3")
    config.coding.enable = True
    config.coding.db_path = str(memory_root / "coding.sqlite3")
+    config.coding.tool_result_max_bytes = coding_tool_result_max_bytes()
    config.recall.episode_recall_enabled = True
    config.recall.episode_summary_enabled = False
    config.recall.dynamic_episode_enabled = True
@@ -1,607 +0,0 @@
-"""Operation-bound read/write capability for the pinned native HY coding store.
-
-HY's coding route writes two layers, experiences and working facts. Rows this
-lease cannot attest stay in the same backend but are invisible here, so they
-never reach a coding prompt and cannot be updated, superseded or linked.
-
-HY's reconcilers catch store exceptions and carry on, so a denial latches and
-recheck() raises it again at finish; a swallowed denial still fails the task.
-Any native method not implemented here is denied rather than passed through.
-"""
-
-from __future__ import annotations
-
-import asyncio
-import json
-from collections.abc import Iterator, Mapping, Sequence
-from copy import deepcopy
-from hashlib import sha256
-from types import MappingProxyType
-from typing import Any, NoReturn
-
-from hy_memory.coding.types import CodingMemory, WorkingFact  # type: ignore[import-untyped]
-
-from one_memory.hy_permission_store import HyNodePermission
-from one_memory.model_processing_permission_guard import (
-    ModelProcessingPermissionLease,
-    permission_error,
-)
-
-_NATIVE_WRITES = frozenset(
-    {
-        "insert",
-        "update",
-        "delete",
-        "insert_fact",
-        "supersede_fact",
-        "set_fact_derived_experiences",
-        "set_experience_source_facts",
-        # Trace rows are raw conversation: writes must be gated like any other.
-        "append_trace",
-    }
-)
-
-_TIMESTAMPS = frozenset({"created_at", "updated_at"})
-
-# SDK 2.0.1 reads `config` off the coding store while routing a coding
-# turn. The wrapper must answer it, but the native `MemoryConfig` carries
-# credentials (llm/embedder api keys, graph and cache passwords) and is a live
-# handle into the backend, so it is never handed out. Only this redacted
-# projection is exposed: credential-bearing keys collapse to None.
-_CONFIG_REDACTED_KEYS = frozenset(
-    {
-        "api_key",
-        "password",
-        "username",
-        "mysql_password",
-        "mysql_user",
-        "extra_headers",
-        "extra_body",
-    }
-)
-
-
-class _RedactedConfig(Mapping[str, Any]):
-    """Read-only config view: readable by key and attribute, never the native one.
-
-    The SDK reaches config values both ways, so this keeps attribute access
-    working. It is deliberately not the native MemoryConfig: it exposes only
-    plain copied values, so no backend handle, connection or callable escapes.
-    """
-
-    __slots__ = ("_values",)
-
-    def __init__(self, values: Mapping[str, Any]) -> None:
-        object.__setattr__(self, "_values", MappingProxyType(dict(values)))
-
-    def __getitem__(self, key: str) -> Any:
-        return self._values[key]
-
-    def __iter__(self) -> Iterator[str]:
-        return iter(self._values)
-
-    def __len__(self) -> int:
-        return len(self._values)
-
-    def __getattr__(self, name: str) -> Any:
-        values = object.__getattribute__(self, "_values")
-        try:
-            return values[name]
-        except KeyError:
-            raise AttributeError(name) from None
-
-    def __setattr__(self, name: str, value: Any) -> NoReturn:
-        raise AttributeError("redacted config is read-only")
-
-    def __repr__(self) -> str:
-        return f"_RedactedConfig({sorted(self._values)})"
-
-
-def _redact_config(value: Any, depth: int = 0) -> Any:
-    """Copy config data into read-only views, collapsing credentials to None.
-
-    The SDK reads this both by attribute and by key, so each config section
-    becomes a mapping-like view rather than a plain dict: attribute access stays
-    available while the native object itself is never handed out.
-    """
-    if depth > 4:
-        return None
-    if value is None or isinstance(value, (str, int, float, bool)):
-        return value
-    if isinstance(value, Mapping):
-        return _RedactedConfig(
-            {
-                key: None if key in _CONFIG_REDACTED_KEYS else _redact_config(item, depth + 1)
-                for key, item in value.items()
-            }
-        )
-    if isinstance(value, (list, tuple)):
-        return [_redact_config(item, depth + 1) for item in value]
-    if hasattr(value, "__dict__") and type(value).__module__.startswith("hy_memory"):
-        return _redact_config(vars(value), depth + 1)
-    return None
-
-
-def _config_projection(native: Any) -> Any:
-    try:
-        return _redact_config(native.config)
-    except Exception:
-        return None
-
-
-def hy_coding_fingerprint(row: CodingMemory | WorkingFact) -> str:
-    """Hash the complete native row as read back, including links and timestamps."""
-    if type(row) is CodingMemory:
-        kind = "experience"
-    elif type(row) is WorkingFact:
-        kind = "fact"
-    else:
-        raise permission_error("hy_coding_store_row_shape")
-    payload = {"kind": kind, **row.to_dict()}
-    return sha256(
-        json.dumps(payload, sort_keys=True, separators=(",", ":"), allow_nan=False).encode()
-    ).hexdigest()
-
-
-def _row_id(row: CodingMemory | WorkingFact) -> str:
-    return str(row.experience_id if type(row) is CodingMemory else row.fact_id)
-
-
-class HyPermissionCodingStore:
-    """Closed, task-exclusive adapter; initialize/close never own the native store.
-
-    node_permissions is the same attested ledger the chat vector store uses;
-    coding rows are keyed by experience or fact id. written_nodes carries the
-    whole lease footprint for each row written here, as the chat store does.
-    """
-
-    def __init__(
-        self,
-        native: Any,
-        lease: ModelProcessingPermissionLease,
-        *,
-        agent_id: str,
-        session_id: str,
-        node_permissions: Mapping[str, HyNodePermission] | None = None,
-    ) -> None:
-        self._native = native
-        self._lease = lease
-        self._agent_id = agent_id
-        self._session_id = session_id
-        self._user_id = lease.user_id
-        self._failure: str | None = None
-        self._closed = False
-        self._lock = asyncio.Lock()
-        self._deleted: set[str] = set()
-        self._written: dict[str, HyNodePermission] = {}
-        self._superseded: dict[str, str] = {}
-        self._evidence_ids = tuple(record.evidence_id for record in lease.records)
-        self._permissions = {
-            key: proof
-            for key, proof in (node_permissions or {}).items()
-            if set(proof.evidence_ids) <= set(self._evidence_ids)
-        }
-        self.recheck()
-        if not agent_id or agent_id == "*" or not session_id or not self._evidence_ids:
-            self._deny("scope")
-
-    def __getattr__(self, name: str) -> NoReturn:
-        if name.startswith("__"):
-            raise AttributeError(name)
-        self._deny("unsupported_operation")
-
-    @property
-    def written_nodes(self) -> Mapping[str, HyNodePermission]:
-        self.recheck()
-        return MappingProxyType(dict(self._written))
-
-    @property
-    def config(self) -> Any:
-        """Read-only, redacted view of the native coding config.
-
-        SDK 2.0.1 reads `config` when it routes a coding turn. Answering
-        it is safe only as a projection: the native MemoryConfig is a live
-        backend handle and carries credentials, so callers get plain values
-        with every credential-bearing key collapsed to None. This carries no
-        identity, permission or tenant state and is not the native object.
-        """
-        self._assert_open()
-        return _config_projection(self._native)
-
-    @property
-    def retired_ids(self) -> frozenset[str]:
-        """Experiences HY deleted in this operation."""
-        self.recheck()
-        return frozenset(self._deleted)
-
-    @property
-    def superseded_ids(self) -> Mapping[str, tuple[str, ...]]:
-        """Facts HY superseded in this operation, with the fact that replaced each.
-
-        A successor linked before it existed must have been written here by
-        now; otherwise the old fact points at a row nobody attested.
-        """
-        self.recheck()
-        if any(new not in self._written for new in self._superseded.values()):
-            self._deny("link_denied")
-        return MappingProxyType({old: (new,) for old, new in self._superseded.items()})
-
-    def recheck(self) -> None:
-        self._assert_open()
-        try:
-            self._lease.recheck()
-        except Exception as error:
-            code = str(error)
-            self._failure = (
-                code
-                if code.startswith("model_content_") and len(code) <= 100
-                else "model_content_hy_coding_store_lease_failed"
-            )
-            raise permission_error(self._failure) from None
-
-    def _assert_open(self) -> None:
-        if self._closed:
-            self._deny("closed")
-        if self._failure:
-            raise permission_error(self._failure)
-
-    async def close(self) -> None:
-        async with self._lock:
-            self._closed = True
-            self._permissions.clear()
-            self._written.clear()
-
-    def _deny(self, suffix: str) -> NoReturn:
-        self._failure = self._failure or f"model_content_hy_coding_store_{suffix}"
-        raise permission_error(self._failure)
-
-    def _scope(self, user_id: str | None) -> None:
-        if user_id not in (None, self._user_id):
-            self._deny("scope")
-
-    async def _invoke(self, method: str, *args: Any, **kwargs: Any) -> Any:
-        writing = method in _NATIVE_WRITES
-        self.recheck() if writing else self._assert_open()
-        try:
-            result = await getattr(self._native, method)(*args, **kwargs)
-        except Exception:
-            self._deny("backend_failed")
-        self.recheck() if writing else self._assert_open()
-        return result
-
-    def _visible(self, row: Any, expected: Any) -> bool:
-        if row is None:
-            return False
-        if type(row) is not expected:
-            self._deny("row_shape")
-        key = _row_id(row)
-        proof = self._permissions.get(key)
-        if proof is None or key in self._deleted or row.user_id != self._user_id:
-            return False
-        if hy_coding_fingerprint(row) != proof.fingerprint:
-            self._deny("row_changed")
-        return True
-
-    async def _experience(self, key: str) -> CodingMemory | None:
-        if key not in self._permissions or key in self._deleted:
-            return None
-        row = await self._invoke("get_by_id", key, user_id=self._user_id)
-        return deepcopy(row) if self._visible(row, CodingMemory) else None
-
-    async def _fact(self, key: str) -> WorkingFact | None:
-        if key not in self._permissions or key in self._deleted:
-            return None
-        rows = await self._invoke("get_facts", [key], user_id=self._user_id)
-        row = rows[0] if rows else None
-        return deepcopy(row) if self._visible(row, WorkingFact) else None
-
-    def _new_row_in_scope(self, row: CodingMemory | WorkingFact) -> bool:
-        return bool(
-            row.user_id == self._user_id
-            and row.agent_id == self._agent_id
-            and row.session_id == self._session_id
-        )
-
-    async def _capture(
-        self, key: str, kind: Any, expected: CodingMemory | WorkingFact | None = None
-    ) -> None:
-        row: Any
-        if kind is CodingMemory:
-            row = await self._invoke("get_by_id", key, user_id=self._user_id)
-        else:
-            rows = await self._invoke("get_facts", [key], user_id=self._user_id)
-            row = rows[0] if rows else None
-        if type(row) is not kind or row.user_id != self._user_id:
-            self._deny("write_mismatch")
-        if expected is not None:
-            stored, planned = row.to_dict(), expected.to_dict()
-            for field in _TIMESTAMPS:
-                stored.pop(field, None)
-                planned.pop(field, None)
-            if stored != planned:
-                self._deny("write_mismatch")
-        proof = HyNodePermission(hy_coding_fingerprint(row), self._evidence_ids)
-        self._permissions[key] = proof
-        self._written[key] = proof
-
-    async def _plan_new(self, row: CodingMemory | WorkingFact, kind: Any) -> None:
-        if type(row) is not kind or not self._new_row_in_scope(row):
-            self._deny("scope")
-        key = _row_id(row)
-        if not key or key in self._permissions or key in self._deleted:
-            self._deny("scope")
-        if kind is CodingMemory:
-            existing = await self._invoke("get_by_id", key)
-        else:
-            existing = next(iter(await self._invoke("get_facts", [key])), None)
-        if existing is not None:
-            self._deny("scope")
-
-    # Reads
-
-    async def get_by_id(
-        self, experience_id: str, user_id: str | None = None
-    ) -> CodingMemory | None:
-        async with self._lock:
-            self._scope(user_id)
-            return await self._experience(experience_id)
-
-    async def get_many(
-        self, experience_ids: list[str], user_id: str | None = None
-    ) -> list[CodingMemory]:
-        async with self._lock:
-            self._scope(user_id)
-            wanted = [
-                key
-                for key in dict.fromkeys(experience_ids)
-                if key in self._permissions and key not in self._deleted
-            ]
-            if not wanted:
-                return []
-            rows = await self._invoke("get_many", wanted, user_id=self._user_id)
-            return [deepcopy(row) for row in rows if self._visible(row, CodingMemory)]
-
-    async def search_by_query_embedding(
-        self,
-        query_embedding: list[float],
-        user_id: str,
-        *,
-        workspace_id: str | None = None,
-        branch: str | None = None,
-        top: int = 40,
-    ) -> list[dict[str, Any]]:
-        async with self._lock:
-            self._scope(user_id)
-            hits = await self._invoke(
-                "search_by_query_embedding",
-                deepcopy(query_embedding),
-                self._user_id,
-                workspace_id=workspace_id,
-                branch=branch,
-                top=top * 4,
-            )
-            result = []
-            for hit in hits:
-                key = hit.get("experience_id")
-                if not isinstance(key, str) or await self._experience(key) is None:
-                    continue
-                result.append(
-                    {
-                        "experience_id": key,
-                        "score": float(hit.get("score", 0.0)),
-                        "matched_key_kind": hit.get("matched_key_kind"),
-                        "matched_key_text": hit.get("matched_key_text"),
-                    }
-                )
-            return result[:top]
-
-    async def get_fact(self, fact_id: str, user_id: str | None = None) -> WorkingFact | None:
-        async with self._lock:
-            self._scope(user_id)
-            return await self._fact(fact_id)
-
-    async def get_facts(self, fact_ids: list[str], user_id: str | None = None) -> list[WorkingFact]:
-        async with self._lock:
-            self._scope(user_id)
-            found = [await self._fact(key) for key in dict.fromkeys(fact_ids)]
-            return [row for row in found if row is not None]
-
-    async def search_facts(
-        self,
-        query_embedding: list[float],
-        user_id: str,
-        *,
-        workspace_id: str | None = None,
-        branch: str | None = None,
-        top: int = 5,
-        query_text: str | None = None,
-        min_score: float = 0.0,
-    ) -> list[tuple[WorkingFact, float]]:
-        async with self._lock:
-            self._scope(user_id)
-            hits = await self._invoke(
-                "search_facts",
-                deepcopy(query_embedding),
-                self._user_id,
-                workspace_id=workspace_id,
-                branch=branch,
-                top=top * 4,
-                query_text=query_text,
-                min_score=min_score,
-            )
-            result = []
-            for row, score in hits:
-                if self._visible(row, WorkingFact):
-                    result.append((deepcopy(row), float(score)))
-            return result[:top]
-
-    async def get_fact_ancestors(
-        self, fact_id: str, *, user_id: str | None = None, max_depth: int = 8
-    ) -> list[WorkingFact]:
-        async with self._lock:
-            self._scope(user_id)
-            if await self._fact(fact_id) is None:
-                return []
-            rows = await self._invoke(
-                "get_fact_ancestors", fact_id, user_id=self._user_id, max_depth=max_depth
-            )
-            # Oldest first; the chain is only as long as its nearest unapproved link.
-            kept: list[WorkingFact] = []
-            for row in reversed(rows):
-                if not self._visible(row, WorkingFact):
-                    break
-                kept.append(deepcopy(row))
-            return list(reversed(kept))
-
-    # Writes
-
-    async def insert(self, memory: CodingMemory) -> None:
-        async with self._lock:
-            self.recheck()
-            await self._plan_new(memory, CodingMemory)
-            await self._invoke("insert", memory)
-            await self._capture(memory.experience_id, CodingMemory, memory)
-
-    async def update(self, memory: CodingMemory) -> None:
-        async with self._lock:
-            self.recheck()
-            if type(memory) is not CodingMemory or not self._new_row_in_scope(memory):
-                self._deny("scope")
-            if await self._experience(memory.experience_id) is None:
-                self._deny("target_denied")
-            await self._invoke("update", memory)
-            await self._capture(memory.experience_id, CodingMemory, memory)
-
-    async def delete(self, experience_id: str, user_id: str) -> bool:
-        async with self._lock:
-            self._scope(user_id)
-            if await self._experience(experience_id) is None:
-                self._deny("target_denied")
-            if await self._invoke("delete", experience_id, user_id=self._user_id) is not True:
-                self._deny("write_failed")
-            if await self._invoke("get_by_id", experience_id) is not None:
-                self._deny("write_mismatch")
-            self._deleted.add(experience_id)
-            self._permissions.pop(experience_id, None)
-            self._written.pop(experience_id, None)
-            return True
-
-    async def insert_fact(self, fact: WorkingFact) -> None:
-        async with self._lock:
-            self.recheck()
-            await self._plan_new(fact, WorkingFact)
-            await self._invoke("insert_fact", fact)
-            await self._capture(fact.fact_id, WorkingFact, fact)
-
-    async def supersede_fact(self, old_fact_id: str, new_fact_id: str) -> bool:
-        async with self._lock:
-            self.recheck()
-            if await self._fact(old_fact_id) is None or old_fact_id == new_fact_id:
-                self._deny("target_denied")
-            # HY links the old fact before it inserts the new one. A successor
-            # that does not exist yet may only become a row this lease writes;
-            # an existing row it cannot attest stays out of reach.
-            if await self._fact(new_fact_id) is None:
-                if not new_fact_id or new_fact_id in self._deleted:
-                    self._deny("target_denied")
-                if next(iter(await self._invoke("get_facts", [new_fact_id])), None) is not None:
-                    self._deny("target_denied")
-            changed = await self._invoke("supersede_fact", old_fact_id, new_fact_id)
-            await self._capture(old_fact_id, WorkingFact)
-            if changed:
-                self._superseded[old_fact_id] = new_fact_id
-            return bool(changed)
-
-    async def set_fact_derived_experiences(self, fact_id: str, experience_ids: list[str]) -> None:
-        async with self._lock:
-            self.recheck()
-            if await self._fact(fact_id) is None:
-                self._deny("target_denied")
-            for key in experience_ids:
-                if await self._experience(key) is None:
-                    self._deny("link_denied")
-            await self._invoke("set_fact_derived_experiences", fact_id, list(experience_ids))
-            await self._capture(fact_id, WorkingFact)
-
-    async def set_experience_source_facts(self, experience_id: str, fact_ids: list[str]) -> None:
-        async with self._lock:
-            self.recheck()
-            if await self._experience(experience_id) is None:
-                self._deny("target_denied")
-            for key in fact_ids:
-                if await self._fact(key) is None:
-                    self._deny("link_denied")
-            await self._invoke("set_experience_source_facts", experience_id, list(fact_ids))
-            await self._capture(experience_id, CodingMemory)
-
-    async def get_recent_trace_messages(
-        self,
-        *,
-        user_id: str,
-        agent_id: str = "default_agent",
-        max_user_turns: int = 5,
-    ) -> list[Any]:
-        """Read recent raw trace rows, still bounded by this lease.
-
-        Public SDK 2.0.1 asks for recent history before a coding write. The native call is
-        scoped to the lease's own user and agent, and rows this lease cannot
-        attest are filtered out, so unapproved conversations never reach a
-        coding prompt through this path.
-        """
-        self._scope(user_id)
-        if agent_id not in (self._agent_id, "*"):
-            self._deny("scope")
-        rows = await self._invoke(
-            "get_recent_trace_messages",
-            user_id=self._user_id,
-            agent_id=self._agent_id,
-            max_user_turns=max_user_turns,
-        )
-        return [row for row in (rows or []) if self._trace_visible(row)]
-
-    async def append_trace(
-        self,
-        messages: Sequence[Any],
-        *,
-        request_id: str,
-        user_id: str,
-        agent_id: str = "default_agent",
-        session_id: str | None = None,
-        workspace_id: str | None = None,
-        branch: str | None = None,
-    ) -> None:
-        """Append raw trace rows for this lease only.
-
-        Trace rows are raw conversation, so this is a write path: the lease is
-        rechecked around the call and every message must belong to the leased
-        user and session. Nothing outside the lease may be recorded.
-        """
-        self._scope(user_id)
-        if agent_id not in (self._agent_id, "*"):
-            self._deny("scope")
-        if session_id not in (None, self._session_id):
-            self._deny("scope")
-        for row in messages or []:
-            if not self._trace_visible(row):
-                self._deny("scope")
-        await self._invoke(
-            "append_trace",
-            list(messages or []),
-            request_id=request_id,
-            user_id=self._user_id,
-            agent_id=self._agent_id,
-            session_id=self._session_id,
-            workspace_id=workspace_id,
-            branch=branch,
-        )
-
-    def _trace_visible(self, row: Any) -> bool:
-        """A trace row may be read or written only inside this lease's scope."""
-        if row is None:
-            return False
-        owner = getattr(row, "user_id", None)
-        session = getattr(row, "session_id", None)
-        return not (
-            (owner is not None and owner != self._user_id)
-            or (session is not None and session != self._session_id)
-        )
@@ -1,582 +0,0 @@
-"""Task-exclusive HY wiring; provenance metadata stays beside the existing Core DB."""
-
-from __future__ import annotations
-
-import asyncio
-import json
-from collections.abc import AsyncIterator, Callable
-from contextlib import asynccontextmanager, nullcontext, suppress
-from pathlib import Path
-from typing import Any, cast
-
-import httpx
-
-from one_memory.database import take_the_write_lock_first
-from one_memory.database_access import main_database
-from one_memory.hy_memory_bridge import (
-    HyMemoryBridge,
-    HyMemoryBridgeError,
-    HyMemoryIngestRequest,
-    HyMemoryIngestResult,
-    HyMemoryOutput,
-    LocalHyEmbedService,
-    _joined_thread,
-)
-from one_memory.hy_permission_coding_store import HyPermissionCodingStore
-from one_memory.hy_permission_store import (
-    HyNodePermission,
-    HyPermissionGraphStore,
-    HyPermissionVectorStore,
-)
-from one_memory.hy_permission_transport import HyPermissionTransport, HyPermissionTransportError
-from one_memory.hy_response_guard import HyResponseGuard
-from one_memory.hy_step_timing import HyStepTimer
-from one_memory.memory_server.migrations import hy_node_permissions_ddl
-from one_memory.memory_tasks import evidence_ref
-from one_memory.model_processing_permission_guard import ModelProcessingPermissionLease
-
-
-class _NativeProofs:
-    def __init__(self, database: Path) -> None:
-        self.database = database
-        self._loaded_states: dict[str, tuple[int, str]] | None = None
-        with main_database(database).write() as db, db:
-            # Schema owned by `memory_server.migrations` (migration 26).
-            db.execute(hy_node_permissions_ddl("sqlite"))
-
-    def load(self, lease: ModelProcessingPermissionLease) -> dict[str, HyNodePermission]:
-        allowed = {record.evidence_id: evidence_ref(record) for record in lease.records}
-        with main_database(self.database).read() as db:
-            rows = db.execute(
-                "SELECT p.node_id,p.fingerprint,p.evidence_json,g.revision,g.status "
-                "FROM hy_node_permissions p LEFT JOIN governed_memories g ON g.memory_id=p.node_id "
-                # Superseded nodes stay attested: every later head links back to them.
-                "WHERE g.status IS NULL OR g.status IN ('confirmed','superseded')"
-            ).fetchall()
-        proofs = {}
-        states = {}
-        for identifier, fingerprint, evidence_json, revision, status in rows:
-            refs = json.loads(evidence_json)
-            if (
-                not isinstance(refs, list)
-                or not refs
-                or not all(isinstance(ref, dict) for ref in refs)
-            ):
-                continue
-            if all(allowed.get(ref.get("evidence_id")) == ref for ref in refs):
-                proofs[identifier] = HyNodePermission(
-                    fingerprint, tuple(ref["evidence_id"] for ref in refs)
-                )
-                if revision is not None:
-                    states[identifier] = (revision, status)
-        self._loaded_states = states
-        return proofs
-
-    def guard(self, values: dict[str, HyNodePermission]) -> Callable[[], None]:
-        identifiers = tuple(values)
-
-        def state() -> dict[str, tuple[int, str]]:
-            found: dict[str, tuple[int, str]] = {}
-            with main_database(self.database).read() as db:
-                for start in range(0, len(identifiers), 900):
-                    batch = identifiers[start : start + 900]
-                    placeholders = ",".join("?" for _ in batch)
-                    for identifier, revision, status in db.execute(
-                        "SELECT memory_id,revision,status FROM governed_memories "
-                        f"WHERE memory_id IN ({placeholders})",
-                        batch,
-                    ):
-                        found[identifier] = (revision, status)
-            return found
-
-        if self._loaded_states is None:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        before = dict(self._loaded_states)
-
-        def check() -> None:
-            if state() != before:
-                raise HyMemoryBridgeError("model_content_context_changed")
-
-        check()
-        return check
-
-    def save(
-        self, lease: ModelProcessingPermissionLease, values: dict[str, HyNodePermission]
-    ) -> None:
-        refs = {record.evidence_id: evidence_ref(record) for record in lease.records}
-        with main_database(self.database).write(autocommit=True) as db:
-            take_the_write_lock_first(db)
-            with db:
-                lease.recheck()
-                db.executemany(
-                    "INSERT INTO hy_node_permissions VALUES (?,?,?) ON CONFLICT(node_id) "
-                    "DO UPDATE SET fingerprint=excluded.fingerprint,"
-                    "evidence_json=excluded.evidence_json",
-                    [
-                        (
-                            identifier,
-                            proof.fingerprint,
-                            json.dumps([refs[key] for key in proof.evidence_ids]),
-                        )
-                        for identifier, proof in values.items()
-                    ],
-                )
-
-    def refresh(
-        self, lease: ModelProcessingPermissionLease, node_id: str
-    ) -> HyNodePermission | None:
-        del lease
-        with main_database(self.database).read() as db:
-            row = db.execute(
-                "SELECT fingerprint,evidence_json FROM hy_node_permissions WHERE node_id=?",
-                (node_id,),
-            ).fetchone()
-        if row is None:
-            return None
-        fingerprint = str(row["fingerprint"])
-        refs = json.loads(row["evidence_json"])
-        if (
-            len(fingerprint) != 64
-            or any(value not in "0123456789abcdef" for value in fingerprint)
-            or not isinstance(refs, list)
-            or not refs
-            or not all(isinstance(ref, dict) for ref in refs)
-        ):
-            return None
-        evidence_ids = tuple(ref.get("evidence_id") for ref in refs)
-        if any(not isinstance(value, str) or not value for value in evidence_ids):
-            return None
-        return HyNodePermission(fingerprint, evidence_ids)
-
-
-class _BoundLease:
-    def __init__(
-        self,
-        lease: ModelProcessingPermissionLease,
-        proof_check: Callable[[], None],
-        step_timer: HyStepTimer | None = None,
-    ) -> None:
-        self.lease = lease
-        self.step_timer = step_timer
-        self.user_id = lease.user_id
-        self.profile = lease.profile
-        self.task_id = lease.task_id
-        self.records = lease.records
-        self.source_context = lease.source_context
-        self.proof_check = proof_check
-        self.vector: HyPermissionVectorStore | None = None
-        self.coding: HyPermissionCodingStore | None = None
-
-    def recheck(self) -> None:
-        timer = self.step_timer
-        with timer.step("permission_check") if timer is not None else nullcontext():
-            self.lease.recheck()
-            self.proof_check()
-            if self.vector is not None:
-                self.vector.recheck()
-            if self.coding is not None:
-                self.coding.recheck()
-
-
-class _Wiring:
-    def __init__(
-        self,
-        client: Any,
-        lease: ModelProcessingPermissionLease,
-        proofs_store: _NativeProofs,
-        proofs: dict[str, HyNodePermission],
-        agent_id: str,
-        session_id: str,
-        transport: httpx.AsyncBaseTransport | None,
-        proof_check: Callable[[], None],
-        step_timer: HyStepTimer,
-    ) -> None:
-        self.client, self.lease, self.proofs_store, self.proofs = (
-            client,
-            lease,
-            proofs_store,
-            proofs,
-        )
-        self.step_timer = step_timer
-        self.agent_id, self.session_id, self.http_transport = agent_id, session_id, transport
-        self.cross_evidence_scope = False
-        self.proof_check = proof_check
-        self.saved: list[tuple[Any, str, Any]] = []
-        self.transport: HyPermissionTransport
-        self.vector: HyPermissionVectorStore
-        self.graph: HyPermissionGraphStore | None = None
-        self.coding: HyPermissionCodingStore
-        self.active = False
-
-    def replace(self, target: Any, name: str, value: Any) -> None:
-        self.saved.append((target, name, getattr(target, name)))
-        setattr(target, name, value)
-
-    async def enter(self) -> None:
-        client, registry = self.client, self.client._registry
-        config = client._config
-        if config.llm.enable_summary or config.llm.enable_query_rewrite:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        bound = _BoundLease(self.lease, self.proof_check, self.step_timer)
-        self.transport = await HyPermissionTransport(
-            bound, http_transport=self.http_transport, step_timer=self.step_timer
-        ).__aenter__()
-        self.active = True
-        try:
-            for guard in getattr(client, "_one_memory_write_guards", ()):
-                if hasattr(type(guard), "step_timer"):
-                    self.replace(guard, "step_timer", self.step_timer)
-            system2 = await registry.get_system2_writer()
-            writers = (await registry.get_writer(), system2._writer)
-            native_vector = client._vector_store
-            store_lease = _BoundLease(self.lease, self.proof_check, self.step_timer)
-            self.vector = HyPermissionVectorStore(
-                native_vector,
-                cast(ModelProcessingPermissionLease, store_lease),
-                agent_id=self.agent_id,
-                session_id=self.session_id,
-                cross_evidence_scope=self.cross_evidence_scope,
-                node_permissions=self.proofs,
-                refresh_permission=lambda node_id: asyncio.to_thread(
-                    self.proofs_store.refresh, self.lease, node_id
-                ),
-                step_timer=self.step_timer,
-            )
-            bound.vector = self.vector
-            embed = client._embed_service
-            if isinstance(embed, LocalHyEmbedService):
-                self.transport.bind_local_embed_service(embed)
-            else:
-                self.transport.bind_embed_service(embed)
-            for writer in writers:
-                provider = writer.mem_agent.llm_provider
-                self.transport.bind_provider(
-                    provider.provider if isinstance(provider, HyResponseGuard) else provider
-                )
-                reconciler = writer.reconciler
-                provider = reconciler._get_llm()
-                self.transport.bind_provider(
-                    provider.provider if isinstance(provider, HyResponseGuard) else provider
-                )
-                self.replace(writer, "_external_vector_store", self.vector)
-                self.replace(writer, "_vector_store", self.vector)
-                self.replace(writer, "_vector_store_initialized", True)
-                self.replace(writer, "_cache", None)
-            self.replace(client, "_vector_store", self.vector)
-            native_graph = client._graph_store
-            if self.cross_evidence_scope and native_graph is not None:
-                self.graph = HyPermissionGraphStore(native_graph, self.proof_check)
-                self.replace(client, "_graph_store", self.graph)
-            else:
-                self.replace(client, "_graph_store", None)
-            self.replace(registry, "_shared_vector_store", self.vector)
-            self.replace(
-                registry,
-                "_shared_graph_store",
-                self.graph if self.graph is not None else None,
-            )
-            self.replace(registry, "_shared_cache", None)
-            self.replace(system2, "_vector_store", self.vector)
-            self.replace(
-                system2,
-                "_graph_store",
-                self.graph if self.graph is not None else None,
-            )
-            self.replace(system2, "_cache", None)
-            # Result fallback may call search; rebuild only its non-generative reader.
-            self.replace(registry, "_reader", None)
-            self.replace(registry, "_readers", {})
-            await self._enter_coding(bound, embed)
-            self.transport.assert_allowed()
-        except BaseException:
-            await self.exit()
-            raise
-
-    async def _enter_coding(self, bound: _BoundLease, embed: Any) -> None:
-        """HY judges and writes the coding route itself; every seam it uses is bound here."""
-        client = self.client
-        # Created now, so the judge and every coding stage share one bound provider.
-        writer = await client._get_coding_writer()
-        stages = (writer, writer.extractor, writer.reconciler, writer.fact_reconciler)
-        native_store = client._coding_store
-        if any(
-            stage.store is not native_store
-            for stage in (writer, writer.reconciler, writer.fact_reconciler)
-        ) or any(
-            item.embed_service is not embed
-            for item in (native_store, writer, writer.reconciler, writer.fact_reconciler)
-        ):
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        providers = {
-            id(p): p for p in (client._coding_llm_provider, *(s.llm_provider for s in stages))
-        }
-        for provider in providers.values():
-            self.transport.bind_provider(provider)
-        self.coding = HyPermissionCodingStore(
-            native_store,
-            cast(
-                ModelProcessingPermissionLease,
-                _BoundLease(self.lease, self.proof_check, self.step_timer),
-            ),
-            agent_id=self.agent_id,
-            session_id=self.session_id,
-            node_permissions=self.proofs,
-        )
-        bound.coding = self.coding
-        for stage in (writer, writer.reconciler, writer.fact_reconciler):
-            self.replace(stage, "store", self.coding)
-        self.replace(client, "_coding_store", self.coding)
-        for stage in (writer, writer.extractor, writer.reconciler):
-            self.replace(stage, "cache", None)
-        # The judge logs its prompt through the client's own cache.
-        self.replace(client, "_cache", None)
-
-    async def finish(self) -> dict[str, HyNodePermission]:
-        self.transport.assert_allowed()
-        self.vector.recheck()
-        self.coding.recheck()
-        return {**self.vector.written_nodes, **self.coding.written_nodes}
-
-    async def retired(self) -> tuple[tuple[str, ...], dict[str, tuple[str, ...]]]:
-        self.transport.assert_allowed()
-        return (
-            tuple(sorted(self.vector.retired_node_ids | self.coding.retired_ids)),
-            {**self.vector.superseded_nodes, **self.coding.superseded_ids},
-        )
-
-    async def exit(self) -> None:
-        if not self.active:
-            return
-        self.active = False
-        try:
-            await self.transport.__aexit__(None, None, None)
-        finally:
-            for target, name, value in reversed(self.saved):
-                setattr(target, name, value)
-            self.saved.clear()
-            if hasattr(self, "vector"):
-                await self.vector.close()
-            if self.graph is not None:
-                await self.graph.close()
-            if hasattr(self, "coding"):
-                await self.coding.close()
-
-
-class _LeasedBridge:
-    def __init__(
-        self,
-        bridge: HyMemoryBridge,
-        lease: ModelProcessingPermissionLease,
-        proofs: _NativeProofs,
-        transport: httpx.AsyncBaseTransport | None,
-    ) -> None:
-        self.bridge, self.lease, self.proofs, self.transport = bridge, lease, proofs, transport
-        self.wiring: _Wiring | None = None
-        self.finished = False
-        self.step_timer = HyStepTimer()
-        self.request_id: str | None = None
-
-    def _step(self, name: str) -> Any:
-        return self.step_timer.step(name)
-
-    async def _recheck(self) -> None:
-        def recheck() -> None:
-            with self._step("permission_check"):
-                self.lease.recheck()
-
-        await _joined_thread(recheck)
-
-    async def _start(self, request: HyMemoryIngestRequest) -> None:
-        if self.wiring is not None or self.finished:
-            raise HyMemoryBridgeError("model_content_context_reused")
-        if request.user_id != self.lease.user_id or set(request.evidence_ids) - {
-            item.evidence_id for item in self.lease.records
-        }:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        await self._start_operation(
-            user_id=request.user_id,
-            agent_id=request.agent_id,
-            request_id=request.request_id,
-            session_id=request.session_id,
-        )
-
-    async def _start_operation(
-        self,
-        *,
-        user_id: str,
-        agent_id: str,
-        request_id: str,
-        session_id: str,
-        cross_evidence_scope: bool = False,
-    ) -> None:
-        if self.wiring is not None or self.finished:
-            raise HyMemoryBridgeError("model_content_context_reused")
-        if user_id != self.lease.user_id:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        self.request_id = request_id
-        await self._recheck()
-        client = cast(Any, await self.bridge._get_client())
-        with self._step("permission_store_read"):
-            known = await asyncio.to_thread(self.proofs.load, self.lease)
-            proof_check = await asyncio.to_thread(self.proofs.guard, known)
-        self.wiring = _Wiring(
-            client,
-            self.lease,
-            self.proofs,
-            known,
-            agent_id,
-            session_id,
-            self.transport,
-            proof_check,
-            self.step_timer,
-        )
-        self.wiring.cross_evidence_scope = cross_evidence_scope
-        await _joined_thread(client._loop_thread.run, self.wiring.enter())
-
-    async def list_agent_ids(self, *, user_id: str, limit: int = 1_000) -> tuple[str, ...]:
-        if user_id != self.lease.user_id or self.wiring is not None or self.finished:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        await self._recheck()
-        result = await self.bridge.list_agent_ids(user_id=user_id, limit=limit)
-        await self._recheck()
-        return result
-
-    async def digest(self, *, user_id: str, agent_id: str) -> dict[str, object]:
-        await self._start_operation(
-            user_id=user_id,
-            agent_id=agent_id,
-            request_id=f"digest:{self.lease.task_id}:{agent_id}",
-            session_id=f"digest:{self.lease.task_id}",
-            cross_evidence_scope=True,
-        )
-        try:
-            result = dict(await self.bridge.digest(user_id=user_id, agent_id=agent_id))
-        except HyMemoryBridgeError:
-            await self._check()
-            raise
-        proofs = await self._check()
-        with self._step("permission_store_write"):
-            await asyncio.to_thread(self.proofs.save, self.lease, proofs)
-        await self.close()
-        await self._recheck()
-        return result
-
-    async def graph_snapshot(self, *, user_id: str, agent_id: str, limit: int = 300) -> Any:
-        await self._start_operation(
-            user_id=user_id,
-            agent_id=agent_id,
-            request_id=f"digest-graph:{self.lease.task_id}:{agent_id}",
-            session_id=f"digest:{self.lease.task_id}",
-            cross_evidence_scope=True,
-        )
-        try:
-            result = await self.bridge.graph_snapshot(
-                user_id=user_id,
-                agent_id=agent_id,
-                limit=limit,
-            )
-        except HyMemoryBridgeError:
-            await self._check()
-            raise
-        await self._check()
-        await self.close()
-        await self._recheck()
-        return result
-
-    async def ingest(self, request: HyMemoryIngestRequest) -> HyMemoryIngestResult:
-        await self._start(request)
-        try:
-            result = await self.bridge.ingest(request)
-        except HyMemoryBridgeError:
-            await self._check()
-            raise
-        return await self._with_retired(result)
-
-    async def ingest_extracted(
-        self, request: HyMemoryIngestRequest, contents: tuple[str, ...]
-    ) -> HyMemoryIngestResult:
-        await self._start(request)
-        try:
-            result = await self.bridge.ingest_extracted(request, contents)
-        except HyMemoryBridgeError:
-            await self._check()
-            raise
-        return await self._with_retired(result)
-
-    async def _with_retired(self, result: HyMemoryIngestResult) -> HyMemoryIngestResult:
-        await self._check()
-        if self.wiring is None:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        try:
-            retired, superseded = await _joined_thread(
-                self.wiring.client._loop_thread.run, self.wiring.retired()
-            )
-        except HyPermissionTransportError as error:
-            await self._recheck()
-            raise HyMemoryBridgeError(error.code) from None
-        return result.model_copy(
-            update={"retired_memory_ids": retired, "superseded_memory_ids": superseded}
-        )
-
-    async def _check(self) -> dict[str, HyNodePermission]:
-        if self.wiring is None or self.finished:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        await self._recheck()
-        try:
-            return await _joined_thread(self.wiring.client._loop_thread.run, self.wiring.finish())
-        except HyPermissionTransportError as error:
-            await self._recheck()
-            raise HyMemoryBridgeError(error.code) from None
-
-    async def profile_memories(self, *, user_id: str, agent_id: str) -> tuple[HyMemoryOutput, ...]:
-        if self.wiring is None or user_id != self.lease.user_id or agent_id != self.wiring.agent_id:
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        result = await self.bridge.profile_memories(user_id=user_id, agent_id=agent_id)
-        proofs = await self._check()
-        with self._step("permission_store_write"):
-            await asyncio.to_thread(self.proofs.save, self.lease, proofs)
-        # SDK context closes before governance can update a previously known memory.
-        await self.close()
-        return result
-
-    async def close(self) -> None:
-        if self.wiring is not None and not self.finished:
-            self.finished = True
-            try:
-                await _joined_thread(self.wiring.client._loop_thread.run, self.wiring.exit())
-            finally:
-                self.step_timer.emit(request_id=self.request_id, task_id=self.lease.task_id)
-
-
-def permission_bridge_factory(
-    resolve_bridge: Callable[..., Any],
-    *,
-    http_transport: httpx.AsyncBaseTransport | None = None,
-) -> Callable[..., Any]:
-    @asynccontextmanager
-    async def factory(
-        lease: ModelProcessingPermissionLease, secret: str | None, data_dir: Path
-    ) -> AsyncIterator[Any]:
-        bridge = resolve_bridge(lease.profile, secret, data_dir)
-        if not isinstance(bridge, HyMemoryBridge):
-            raise HyMemoryBridgeError("model_content_context_unverified")
-        proofs = _NativeProofs(data_dir / "one-memory.sqlite3")
-        async with bridge.exclusive_access():
-            await bridge.prepare()
-            bound = _LeasedBridge(bridge, lease, proofs, http_transport)
-            try:
-                yield bound
-            except BaseException as error:
-                with suppress(Exception):
-                    await bound.close()
-                if isinstance(error, HyPermissionTransportError):
-                    await _joined_thread(lease.recheck)
-                    raise HyMemoryBridgeError(error.code) from None
-                raise
-            else:
-                try:
-                    await bound.close()
-                except HyPermissionTransportError as error:
-                    raise HyMemoryBridgeError(error.code) from None
-
-    return factory
@@ -1,880 +0,0 @@
-"""Operation-bound read/write capability for the pinned native HY chat vector store.
-
-No semantic storage or extraction lives here. Pre-existing nodes require a full
-native-node attestation with complete approved Evidence provenance; known content
-alone cannot authorize tags, custom fields, vectors or links. Unknown nodes stay
-in the same HY backend but are not exposed to this operation.
-
-The caller must exclusively own the native backend's writes during execution:
-its API has no compare-and-swap, and a swallowed get_by_id error can look missing.
-This component cannot make that check/upsert interval atomic. Factory integration
-must replace client/registry/both writers/cached-reader references, bind physical
-LLM/embedding sends to the lease AND this denial latch, and guard publication.
-Coding rows go through hy_permission_coding_store instead; graph, entities,
-digest, sweeper and summary remain unsupported.
-"""
-
-from __future__ import annotations
-
-import asyncio
-import json
-import math
-import struct
-import weakref
-from collections.abc import AsyncIterator, Awaitable, Callable, Mapping
-from contextlib import asynccontextmanager, nullcontext
-from copy import deepcopy
-from dataclasses import dataclass
-from hashlib import sha256
-from types import MappingProxyType
-from typing import Any, NoReturn
-
-from hy_memory.models.memory import (  # type: ignore[import-untyped]
-    MemoryLayer,
-    MemoryNode,
-    MemoryStatus,
-)
-
-from one_memory.hy_step_timing import HyStepTimer
-from one_memory.model_processing_permission_guard import (
-    ModelProcessingPermissionLease,
-    permission_error,
-)
-
-_WRITE_LOCKS: weakref.WeakKeyDictionary[asyncio.AbstractEventLoop, asyncio.Lock] = (
-    weakref.WeakKeyDictionary()
-)
-
-
-def _native_write_lock() -> asyncio.Lock:
-    loop = asyncio.get_running_loop()
-    lock = _WRITE_LOCKS.get(loop)
-    if lock is None:
-        lock = asyncio.Lock()
-        _WRITE_LOCKS[loop] = lock
-    return lock
-
-
-@asynccontextmanager
-async def native_store_read_slot() -> AsyncIterator[bool]:
-    """Admit one native HY read only while no write is active.
-
-    Chroma's native reader may retain the GIL after an asyncio timeout. Running
-    it concurrently with HY's multi-step store writes has deadlocked the Core
-    under sustained imports. Reads therefore share the physical store lock,
-    but never wait behind a writer: host recall can immediately use the local
-    governance ledger instead.
-    """
-    lock = _native_write_lock()
-    if lock.locked():
-        yield False
-        return
-    await lock.acquire()
-    try:
-        yield True
-    finally:
-        lock.release()
-
-
-@dataclass(frozen=True)
-class HyNodePermission:
-    """Trusted caller's full-node attestation, not permission inferred from an ID."""
-
-    fingerprint: str
-    evidence_ids: tuple[str, ...]
-
-
-def hy_node_fingerprint(node: MemoryNode) -> str:
-    """Hash the complete native serialized node, including vectors and metadata.
-
-    Timestamp precision follows HY's to_dict contract (integer Unix seconds).
-    Attest the node read from the backend, not an unpersisted input object.
-    """
-    if type(node) is not MemoryNode:
-        raise permission_error("hy_store_node_shape")
-    return sha256(
-        json.dumps(node.to_dict(), sort_keys=True, separators=(",", ":"), allow_nan=False).encode()
-    ).hexdigest()
-
-
-_CHAT_LAYERS = frozenset(
-    {
-        MemoryLayer.L0_BASIC_INFO,
-        MemoryLayer.L1_RAW,
-        MemoryLayer.L2_FACT,
-        MemoryLayer.L4_IDENTITY,
-        MemoryLayer.L7_INTENTION,
-    }
-)
-_UPDATE_FIELDS = frozenset(
-    {
-        "content",
-        "embedding",
-        "tags",
-        "tag_ners",
-        "layer",
-        "memory_at",
-        "gmt_modified",
-        "is_latest",
-        "status",
-        "supersedes",
-        "superseded_by",
-        "source_raw_memory_id",
-        # SDK 2.0.1 sends both of these on an in-place reconcile UPDATE.
-        # They are metadata only and must never carry identity, permission or
-        # tenant state: `owner` is a provenance label whose only values are
-        # "user" / "agent" (who said this line), and `custom` is the SDK's own
-        # free-form notes field. Neither may hold a user_id, scope, grant or
-        # tenancy marker, so allowing the field names does not widen the
-        # cross-user boundary that the store still enforces per node.
-        "owner",
-        "custom",
-    }
-)
-_DEDUP_AUDIT_FIELDS = frozenset({"shadow_reason", "dedup_kept_id"})
-_DIGEST_COUNTER_FIELD = "s2_evidence_count"
-
-
-def _write_view(node: MemoryNode) -> dict[str, Any]:
-    """Only native Chroma's empty-string and float32 storage normalization."""
-    payload: dict[str, Any] = node.to_dict()
-    payload["source_session_id"] = payload["source_session_id"] or ""
-    # The pinned public SDK serializes tag_ners as JSON, but its Chroma read path omits this
-    # field from the JSON decode list. Decode only the exact list-of-strings
-    # shape that the write path produced; malformed values still fail the
-    # complete-node comparison below.
-    tag_ners = payload.get("tag_ners")
-    if (
-        isinstance(tag_ners, list)
-        and tag_ners
-        and all(isinstance(value, str) and len(value) == 1 for value in tag_ners)
-        and tag_ners[0] == "["
-        and tag_ners[-1] == "]"
-    ):
-        tag_ners = "".join(tag_ners)
-    if isinstance(tag_ners, str):
-        try:
-            decoded = json.loads(tag_ners)
-        except (json.JSONDecodeError, TypeError):
-            decoded = None
-        if isinstance(decoded, list) and all(isinstance(value, str) for value in decoded):
-            payload["tag_ners"] = decoded
-    if payload.get("tag_ners") in ([], ["[", "]"]):
-        payload["tag_ners"] = []
-    # HY Chroma encodes None as ""; these JSON fields keep "" when json.loads
-    # fails (vector_store_chroma.py:741-788). Only an absent link is equivalent.
-    for field in ("supersedes", "superseded_by"):
-        if payload[field] == "":
-            payload[field] = None
-    if payload["embedding"] is not None:
-        payload["embedding"] = [
-            struct.unpack("f", struct.pack("f", n))[0] for n in payload["embedding"]
-        ]
-        if not all(math.isfinite(n) for n in payload["embedding"]):
-            raise ValueError("nonfinite_embedding")
-    return payload
-
-
-def _same_written_vector(actual: list[float] | None, expected: list[float] | None) -> bool:
-    if actual == expected:
-        return True
-    if actual is None or expected is None or len(actual) != len(expected):
-        return False
-    # Chroma's cosine index can renormalize float32 vectors on readback. Bound
-    # only this write roundtrip; the saved full-node proof uses exact readback.
-    return all(
-        math.isclose(left, right, rel_tol=4 * 2**-23, abs_tol=0.0)
-        for left, right in zip(actual, expected, strict=True)
-    )
-
-
-_NATIVE_WRITES = frozenset(
-    {"upsert", "upsert_batch", "update_payload", "update_embedding", "delete"}
-)
-
-
-class HyPermissionGraphStore:
-    """Lease-gated graph store for HY System 2's digest transaction.
-
-    The native graph is deliberately absent from ordinary ingest leases. Digest
-    needs it to write L6 schemas, but every operation must remain inside the
-    same frozen permission lease rather than exposing the raw store.
-    """
-
-    def __init__(self, native: Any, recheck: Any) -> None:
-        self._native = native
-        self._recheck = recheck
-        self._closed = False
-
-    def recheck(self) -> None:
-        if self._closed:
-            raise permission_error("hy_graph_store_closed")
-        self._recheck()
-
-    async def close(self) -> None:
-        self._closed = True
-
-    def __getattr__(self, name: str) -> Any:
-        if name.startswith("_") and name != "_run":
-            raise AttributeError(name)
-        attribute = getattr(self._native, name)
-        if not callable(attribute):
-            return attribute
-
-        async def guarded(*args: Any, **kwargs: Any) -> Any:
-            self.recheck()
-            result = await attribute(*args, **kwargs)
-            self.recheck()
-            return result
-
-        return guarded
-
-
-class HyPermissionVectorStore:
-    """Closed, task-exclusive adapter; initialize/close never own the native store.
-
-    node_permissions may attest additional native raw/profile nodes not represented
-    in known_memories. Every attestation must cover all its contributors, and all
-    Evidence IDs must already be in this lease. Known entries also constrain exact
-    content/owner. Without a full attestation even a known entry is denied.
-
-    written_nodes records successful writes with the entire lease footprint, a
-    conservative provenance bound, not a claim of semantic citation accuracy.
-    These transient fingerprints must be persisted by the integrating factory if
-    future operations are to reuse raw/profile nodes. No content is stored here.
-    """
-
-    __slots__ = (
-        "__native",
-        "__lease",
-        "__agent_id",
-        "__session_id",
-        "__cross_evidence_scope",
-        "__user_id",
-        "__known",
-        "__permissions",
-        "__written",
-        "__evidence_ids",
-        "__deleted",
-        "__shadowed",
-        "__superseded",
-        "__heads",
-        "__lock",
-        "__closed",
-        "__failure",
-        "__step_timer",
-        "__refresh_permission",
-    )
-
-    def __init__(
-        self,
-        native: Any,
-        lease: ModelProcessingPermissionLease,
-        *,
-        agent_id: str,
-        session_id: str,
-        cross_evidence_scope: bool = False,
-        node_permissions: Mapping[str, HyNodePermission] | None = None,
-        refresh_permission: Callable[[str], Awaitable[HyNodePermission | None]] | None = None,
-        step_timer: HyStepTimer | None = None,
-    ) -> None:
-        self.__step_timer = step_timer
-        self.__native = native
-        self.__lease = lease
-        self.__agent_id = agent_id
-        self.__session_id = session_id
-        self.__cross_evidence_scope = cross_evidence_scope
-        self.__user_id = lease.user_id
-        self.__failure: str | None = None
-        self.__closed = False
-        self.__lock = asyncio.Lock()
-        self.__deleted: set[str] = set()
-        self.__written: dict[str, HyNodePermission] = {}
-        self.__shadowed: set[str] = set()
-        self.__superseded: dict[str, tuple[str, ...]] = {}
-        self.__heads: dict[str, tuple[str, ...]] = {}
-        self.__refresh_permission = refresh_permission
-        self.recheck()
-        self.__evidence_ids = tuple(record.evidence_id for record in lease.records)
-        current = lease.source_context.current if lease.source_context else lease.records
-        conversation_mismatch = any(r.conversation_id != session_id for r in current)
-        if (
-            not lease.task_id
-            or not agent_id
-            or agent_id == "*"
-            or not session_id
-            or not current
-            or (conversation_mismatch and not cross_evidence_scope)
-            or not self.__evidence_ids
-            or len(set(self.__evidence_ids)) != len(lease.records)
-            or any(r.user_id != self.__user_id for r in lease.records)
-        ):
-            self.__deny("scope")
-        known = lease.source_context.known_memories if lease.source_context else ()
-        self.__known = {item.hy_memory_id: item for item in known}
-        if len(self.__known) != len(known):
-            self.__deny("provenance")
-        self.__permissions = dict(node_permissions or {})
-        for key, proof in self.__permissions.items():
-            if (
-                not key
-                or not isinstance(proof, HyNodePermission)
-                or len(proof.fingerprint) != 64
-                or any(c not in "0123456789abcdef" for c in proof.fingerprint)
-                or not isinstance(proof.evidence_ids, tuple)
-                or not proof.evidence_ids
-                or len(set(proof.evidence_ids)) != len(proof.evidence_ids)
-                or not set(proof.evidence_ids) <= set(self.__evidence_ids)
-            ):
-                self.__deny("provenance")
-        for key, item in self.__known.items():
-            if not item.evidence_ids or not set(item.evidence_ids) <= set(self.__evidence_ids):
-                self.__deny("provenance")
-            known_proof = self.__permissions.get(key)
-            if known_proof and not set(item.evidence_ids) <= set(known_proof.evidence_ids):
-                self.__deny("provenance")
-
-    @property
-    def written_nodes(self) -> Mapping[str, HyNodePermission]:
-        """Latest verified write proofs; each access returns a new immutable snapshot.
-
-        Export after SDK writes have drained and before close/governance projection.
-        Merge these rows into the lineage ledger, then remove deleted_node_ids;
-        this write delta does not include untouched pre-existing permissions.
-        """
-        self.recheck()
-        return MappingProxyType(dict(self.__written))
-
-    @property
-    def deleted_node_ids(self) -> frozenset[str]:
-        """Verified native deletions in this operation, for lineage-ledger cleanup."""
-        self.recheck()
-        return frozenset(self.__deleted)
-
-    @property
-    def retired_node_ids(self) -> frozenset[str]:
-        """Nodes HY deleted or shadowed (its logical delete) in this operation."""
-        self.recheck()
-        return frozenset(self.__deleted | self.__shadowed)
-
-    @property
-    def superseded_nodes(self) -> Mapping[str, tuple[str, ...]]:
-        """Nodes HY marked superseded in this operation, with what replaced them.
-
-        HY writes the two directions separately and may fail either one, so a
-        replacement counts whether it was recorded on the old node or only on
-        the new head.
-        """
-        self.recheck()
-        result = {}
-        for key, by in self.__superseded.items():
-            heads = set(by) | {head for head, olds in self.__heads.items() if key in olds}
-            heads -= self.__deleted | {key}
-            if key not in self.__deleted and heads:
-                result[key] = tuple(sorted(heads))
-        return MappingProxyType(result)
-
-    @property
-    def supports_fulltext(self) -> bool:
-        return False
-
-    def recheck(self) -> None:
-        """Must also gate transport/publication after HY swallows a store failure."""
-        self.__assert_open()
-        try:
-            self.__lease.recheck()
-        except Exception as error:
-            code = str(error)
-            self.__failure = (
-                code
-                if code.startswith("model_content_") and len(code) <= 100
-                else "model_content_hy_store_lease_failed"
-            )
-            raise permission_error(self.__failure) from None
-
-    def __assert_open(self) -> None:
-        if self.__closed:
-            self.__deny("closed")
-        if self.__failure:
-            raise permission_error(self.__failure)
-
-    async def initialize(self) -> None:
-        self.recheck()
-
-    async def close(self) -> None:
-        """Revoke this wrapper locally without rechecking context or closing HY.
-
-        Cleanup must remain possible after governance projection changes the
-        lease's known-memory revisions, or after permission has been revoked.
-        """
-        async with self.__lock:
-            self.__closed = True
-            self.__permissions.clear()
-            self.__written.clear()
-
-    async def has_tag_embedding(self, user_id: str, tag: str) -> bool:
-        self.recheck()
-        # Native chat calls this best-effort helper. NotImplemented prevents it
-        # from embedding tags or falling through to an unguarded entity store.
-        raise NotImplementedError("hy_permission_store_tag_index_unsupported")
-
-    def __deny(self, suffix: str) -> NoReturn:
-        self.__failure = self.__failure or f"model_content_hy_store_{suffix}"
-        raise permission_error(self.__failure)
-
-    async def __invoke(self, method: str, *args: Any, **kwargs: Any) -> Any:
-        writing = method in _NATIVE_WRITES
-        self.recheck() if writing else self.__assert_open()
-        step = "vector_store_write" if writing else "vector_store_read"
-        first = args[0] if args else None
-        try:
-            with (
-                self.__step_timer.step(step, items=len(first) if isinstance(first, list) else 1)
-                if self.__step_timer is not None
-                else nullcontext()
-            ):
-                result = await getattr(self.__native, method)(*args, **kwargs)
-        except Exception:
-            self.__deny("backend_failed")
-        self.recheck() if writing else self.__assert_open()
-        return result
-
-    async def __read_native(self, key: str) -> MemoryNode | None:
-        value = await self.__invoke("get_by_id", key)
-        if value is not None and (type(value) is not MemoryNode or value.node_id != key):
-            self.__deny("node_shape")
-        return deepcopy(value)
-
-    def __fingerprint(self, node: MemoryNode) -> str:
-        try:
-            return hy_node_fingerprint(node)
-        except (TypeError, ValueError, AttributeError, OverflowError):
-            self.__deny("node_shape")
-
-    def __write_view(self, node: MemoryNode) -> dict[str, Any]:
-        self.__fingerprint(node)
-        try:
-            return _write_view(node)
-        except (TypeError, ValueError, OverflowError, struct.error):
-            self.__deny("node_shape")
-
-    def __in_scope(self, node: MemoryNode) -> bool:
-        if node.user_id != self.__user_id or node.layer not in _CHAT_LAYERS:
-            return False
-        if node.layer is MemoryLayer.L0_BASIC_INFO:
-            # Public SDK 2.0.1 migrates verified legacy agent-level profiles into one
-            # user-level chain. Existing nodes still require a full proof in
-            # __load; accepting their historical agent id does not authorize
-            # unproven cross-agent memories.
-            return True
-        if self.__cross_evidence_scope:
-            return bool(node.agent_id == self.__agent_id)
-        return bool(node.agent_id == self.__agent_id and node.session_id == self.__session_id)
-
-    def __new_node_in_scope(self, node: MemoryNode) -> bool:
-        if not self.__in_scope(node):
-            return False
-        if node.layer is MemoryLayer.L0_BASIC_INFO:
-            return node.agent_id in ("", self.__agent_id) and node.session_id in (
-                "",
-                self.__session_id,
-            )
-        return True
-
-    def __references(self, node: MemoryNode) -> list[str]:
-        refs = []
-        for field in ("supersedes", "superseded_by", "evidence_chain"):
-            values = getattr(node, field)
-            if field != "evidence_chain" and values == "":
-                continue
-            if values is not None:
-                if not isinstance(values, list) or any(
-                    not isinstance(v, str) or not v for v in values
-                ):
-                    self.__deny("node_shape")
-                refs.extend(values)
-        if node.source_raw_memory_id is not None:
-            if not isinstance(node.source_raw_memory_id, str) or not node.source_raw_memory_id:
-                self.__deny("node_shape")
-            refs.append(node.source_raw_memory_id)
-        return refs
-
-    async def __load(
-        self,
-        key: str,
-        first: MemoryNode | None = None,
-        *,
-        allow_deleted_links: bool = False,
-    ) -> MemoryNode | None:
-        pending = [key]
-        checked: dict[str, MemoryNode] = {}
-        while pending:
-            current = pending.pop()
-            if current in checked:
-                continue
-            # HY dedup deletes head then ancestors. Only this operation's
-            # verified tombstones may terminate a closure during deletion.
-            if allow_deleted_links and current != key and current in self.__deleted:
-                continue
-            proof = self.__permissions.get(current)
-            if proof is None or current in self.__deleted:
-                return None
-            value = (
-                first if current == key and first is not None else await self.__read_native(current)
-            )
-            if value is None or not self.__in_scope(value):
-                return None
-            if self.__fingerprint(value) != proof.fingerprint:
-                refreshed = (
-                    await self.__refresh_permission(current)
-                    if self.__refresh_permission is not None
-                    else None
-                )
-                if refreshed is None or self.__fingerprint(value) != refreshed.fingerprint:
-                    self.__deny("node_changed")
-                self.__permissions[current] = refreshed
-                proof = refreshed
-            known = self.__known.get(current) if current not in self.__written else None
-            if known and (value.content != known.content or value.owner != known.owner):
-                self.__deny("node_changed")
-            checked[current] = value
-            pending.extend(self.__references(value))
-        return deepcopy(checked[key])
-
-    async def get_by_id(self, node_id: str) -> MemoryNode | None:
-        async with self.__lock:
-            self.recheck()
-            result = await self.__load(node_id)
-            self.recheck()
-            return result
-
-    async def get_by_ids(self, node_ids: list[str]) -> list[MemoryNode]:
-        async with self.__lock:
-            self.recheck()
-            result = [
-                n for key in dict.fromkeys(node_ids) if (n := await self.__load(key)) is not None
-            ]
-            self.recheck()
-            return result
-
-    async def get_embeddings(self, node_ids: list[str]) -> dict[str, list[float]]:
-        return {
-            n.node_id: list(n.embedding)
-            for n in await self.get_by_ids(node_ids)
-            if n.embedding is not None
-        }
-
-    def __scope(self, user_id: str | None, agent_id: str | None) -> None:
-        if user_id not in (None, self.__user_id) or agent_id not in (None, self.__agent_id):
-            self.__deny("scope")
-
-    async def list_by_user(
-        self,
-        user_id: str | None = None,
-        agent_id: str | None = None,
-        limit: int = 10000,
-        status_filter: list[Any] | None = None,
-        layers: list[Any] | None = None,
-        user_ids: list[str] | None = None,
-        agent_ids: list[str] | None = None,
-        only_latest: bool = False,
-        with_vectors: bool = True,
-        allow_raw_layer: bool = False,
-    ) -> list[MemoryNode]:
-        async with self.__lock:
-            self.recheck()
-            effective_users = user_ids or ([user_id] if user_id else [])
-            effective_agents = agent_ids or ([agent_id] if agent_id else [])
-            if effective_users != [self.__user_id] or any(
-                value != self.__agent_id for value in effective_agents
-            ):
-                self.__deny("scope")
-            rows = await self.__invoke(
-                "list_by_user",
-                user_id=self.__user_id,
-                agent_id=agent_id,
-                limit=limit,
-                status_filter=status_filter,
-                layers=layers,
-                user_ids=[self.__user_id],
-                agent_ids=effective_agents or None,
-                only_latest=only_latest,
-                with_vectors=with_vectors,
-                allow_raw_layer=allow_raw_layer,
-            )
-            result = []
-            for row in rows:
-                if type(row) is not MemoryNode:
-                    self.__deny("node_shape")
-                value = await self.__load(row.node_id)
-                if (
-                    value is not None
-                    and (not layers or value.layer in layers or value.layer.value in layers)
-                    and (
-                        not status_filter
-                        or value.status in status_filter
-                        or value.status.value in status_filter
-                    )
-                ):
-                    result.append(value)
-            self.recheck()
-            return result[:limit]
-
-    async def search(
-        self,
-        query_embedding: list[float],
-        isolation_key: str = "",
-        isolation_keys: list[str] | None = None,
-        user_id: str | None = None,
-        user_ids: list[str] | None = None,
-        agent_ids: list[str] | None = None,
-        layers: list[Any] | None = None,
-        limit: int = 10,
-        score_threshold: float = 0.0,
-        status_filter: list[Any] | None = None,
-        only_latest: bool = True,
-        tags_match_any: list[str] | None = None,
-        created_after: float | None = None,
-    ) -> list[dict[str, Any]]:
-        async with self.__lock:
-            self.recheck()
-            self.__scope(user_id, None)
-            key = MemoryNode.build_isolation_key(self.__user_id, self.__agent_id, self.__session_id)
-            if (
-                any(uid != self.__user_id for uid in user_ids or [])
-                or any(aid != self.__agent_id for aid in agent_ids or [])
-                or isolation_key not in ("", key)
-                or any(ik != key for ik in isolation_keys or [])
-            ):
-                self.__deny("scope")
-            hits = await self.__invoke(
-                "search",
-                query_embedding=deepcopy(query_embedding),
-                isolation_key=isolation_key,
-                isolation_keys=isolation_keys,
-                user_id=self.__user_id,
-                user_ids=[self.__user_id],
-                agent_ids=[self.__agent_id],
-                layers=layers,
-                limit=limit,
-                score_threshold=score_threshold,
-                status_filter=status_filter,
-                only_latest=only_latest,
-                tags_match_any=tags_match_any,
-                created_after=created_after,
-            )
-            result = []
-            for hit in hits:
-                value = await self.__load(hit["node_id"])
-                if value is None:
-                    continue
-                score = float(hit["score"])
-                if not math.isfinite(score):
-                    self.__deny("node_shape")
-                # Do not forward arbitrary backend hit metadata. Rehydrate the
-                # full fingerprinted node; native search often omits embedding.
-                result.append({"node_id": value.node_id, "node": value, "score": score})
-            self.recheck()
-            return result[:limit]
-
-    async def __target(
-        self,
-        key: str,
-        first: MemoryNode | None = None,
-        *,
-        allow_deleted_links: bool = False,
-    ) -> MemoryNode:
-        value = await self.__load(key, first, allow_deleted_links=allow_deleted_links)
-        if value is None:
-            self.__deny("target_denied")
-        return value
-
-    async def __check_links(self, value: MemoryNode) -> None:
-        for ref in self.__references(value):
-            if await self.__load(ref) is None:
-                self.__deny("link_denied")
-
-    async def __plan_upsert(self, value: MemoryNode) -> MemoryNode:
-        value = deepcopy(value)
-        if type(value) is not MemoryNode or not value.node_id or not self.__in_scope(value):
-            self.__deny("scope")
-        existing = await self.__read_native(value.node_id)
-        if existing is not None:
-            await self.__target(value.node_id, existing)
-        elif (
-            not self.__new_node_in_scope(value)
-            or value.node_id in self.__permissions
-            or value.node_id in self.__known
-            or value.node_id in self.__deleted
-        ):
-            self.__deny("scope")
-        await self.__check_links(value)
-        self.__write_view(value)
-        return value
-
-    async def __capture(
-        self,
-        expected: MemoryNode,
-        *,
-        allow_native_modified_time: bool = False,
-    ) -> None:
-        actual = await self.__read_native(expected.node_id)
-        if actual is None:
-            self.__deny("write_mismatch")
-        actual_view, expected_view = self.__write_view(actual), self.__write_view(expected)
-        if allow_native_modified_time:
-            if actual.gmt_modified < expected.gmt_modified:
-                self.__deny("write_mismatch")
-            actual_view["gmt_modified"] = expected_view["gmt_modified"]
-        actual_vector = actual_view.pop("embedding")
-        expected_vector = expected_view.pop("embedding")
-        if actual_view != expected_view or not _same_written_vector(actual_vector, expected_vector):
-            self.__deny("write_mismatch")
-        proof = HyNodePermission(self.__fingerprint(actual), self.__evidence_ids)
-        self.__permissions[actual.node_id] = proof
-        self.__written[actual.node_id] = proof
-        status = getattr(actual.status, "value", actual.status)
-        if status == MemoryStatus.SHADOW.value:
-            self.__shadowed.add(actual.node_id)
-        else:
-            self.__shadowed.discard(actual.node_id)
-        if status == MemoryStatus.SUPERSEDED.value:
-            by = actual.superseded_by
-            self.__superseded[actual.node_id] = tuple(by) if isinstance(by, list) else ()
-        else:
-            self.__superseded.pop(actual.node_id, None)
-        olds = actual.supersedes
-        if isinstance(olds, list) and olds:
-            self.__heads[actual.node_id] = tuple(olds)
-        else:
-            self.__heads.pop(actual.node_id, None)
-
-    async def upsert(self, node: MemoryNode) -> str:
-        async with _native_write_lock(), self.__lock:
-            self.recheck()
-            expected = await self.__plan_upsert(node)
-            result = await self.__invoke("upsert", deepcopy(expected))
-            if result != expected.node_id:
-                self.__deny("write_mismatch")
-            await self.__capture(expected)
-            return str(result)
-
-    async def upsert_batch(self, nodes: list[MemoryNode]) -> list[str]:
-        async with _native_write_lock(), self.__lock:
-            self.recheck()
-            if len({n.node_id for n in nodes}) != len(nodes):
-                self.__deny("node_shape")
-            # No forward links to uncommitted batch members. Preflight every
-            # existing target before the backend can make a partial write.
-            planned = [await self.__plan_upsert(n) for n in nodes]
-            ids = [n.node_id for n in planned]
-            result = await self.__invoke("upsert_batch", deepcopy(planned))
-            if result != ids:
-                self.__deny("write_mismatch")
-            for expected in planned:
-                await self.__capture(expected)
-            return ids
-
-    async def update_payload(self, node_id: str, updates: dict[str, Any]) -> bool:
-        async with _native_write_lock(), self.__lock:
-            self.recheck()
-            current = await self.__target(node_id)
-            updates = deepcopy(updates)
-            dedup_audit = False
-            if set(updates) & _DEDUP_AUDIT_FIELDS:
-                if (
-                    not set(updates) <= (_UPDATE_FIELDS | _DEDUP_AUDIT_FIELDS)
-                    or updates.get("status") != MemoryStatus.SHADOW.value
-                    or updates.get("is_latest") is not False
-                    or not isinstance(updates.get("shadow_reason"), str)
-                    or not updates["shadow_reason"].startswith("dedup:")
-                    or not isinstance(updates.get("dedup_kept_id"), str)
-                    or not updates["dedup_kept_id"]
-                ):
-                    self.__deny("update_fields")
-                await self.__target(updates["dedup_kept_id"])
-                custom = deepcopy(current.custom)
-                custom["hy_dedup"] = {
-                    "reason": updates.pop("shadow_reason"),
-                    "kept_id": updates.pop("dedup_kept_id"),
-                }
-                updates["custom"] = custom
-                dedup_audit = True
-            allowed = _UPDATE_FIELDS | ({"custom"} if dedup_audit else set())
-            if self.__cross_evidence_scope and set(updates) == {"custom"}:
-                custom = updates.get("custom")
-                if not isinstance(custom, dict):
-                    self.__deny("update_fields")
-                current_custom = current.custom if isinstance(current.custom, dict) else {}
-                previous = current_custom.get(_DIGEST_COUNTER_FIELD, 0)
-                proposed = custom.get(_DIGEST_COUNTER_FIELD)
-                if (
-                    not isinstance(previous, int)
-                    or isinstance(previous, bool)
-                    or not isinstance(proposed, int)
-                    or isinstance(proposed, bool)
-                    or proposed != previous + 1
-                    or set(custom) != set(current_custom) | {_DIGEST_COUNTER_FIELD}
-                    or any(custom[key] != value for key, value in current_custom.items())
-                ):
-                    self.__deny("update_fields")
-                allowed |= {"custom"}
-            # `custom` is an allowed field name (SDK 2.0.1 sends it on a
-            # reconcile UPDATE), but the System 2 evidence counter it can hold
-            # stays digest-only: only the digest scope may advance it, and only
-            # by one per write. Letting the name through must not let an
-            # ordinary write move that counter.
-            if not self.__cross_evidence_scope and _DIGEST_COUNTER_FIELD in (
-                updates.get("custom") or {}
-            ):
-                self.__deny("update_fields")
-            if not set(updates) <= allowed:
-                self.__deny("update_fields")
-            try:
-                if "layer" in updates:
-                    MemoryLayer(updates["layer"])
-                if "status" in updates:
-                    MemoryStatus(updates["status"])
-                if "tag_ners" in updates and (
-                    not isinstance(updates["tag_ners"], list)
-                    or any(not isinstance(value, str) for value in updates["tag_ners"])
-                ):
-                    self.__deny("node_shape")
-                expected = MemoryNode.from_dict({**current.to_dict(), **updates})
-                self.__fingerprint(expected)
-            except (TypeError, ValueError):
-                self.__deny("node_shape")
-            if not self.__in_scope(expected):
-                self.__deny("scope")
-            await self.__check_links(expected)
-            self.__write_view(expected)
-            if await self.__invoke("update_payload", node_id, updates) is not True:
-                self.__deny("write_failed")
-            await self.__capture(expected, allow_native_modified_time=True)
-            return True
-
-    async def update_embedding(self, node_id: str, embedding: list[float]) -> bool:
-        async with _native_write_lock(), self.__lock:
-            self.recheck()
-            expected = await self.__target(node_id)
-            expected.embedding = deepcopy(embedding)
-            self.__write_view(expected)
-            if await self.__invoke("update_embedding", node_id, deepcopy(embedding)) is not True:
-                self.__deny("write_failed")
-            await self.__capture(expected, allow_native_modified_time=True)
-            return True
-
-    async def delete(self, node_id: str) -> bool:
-        async with _native_write_lock(), self.__lock:
-            self.recheck()
-            await self.__target(node_id, allow_deleted_links=True)
-            if await self.__invoke("delete", node_id) is not True:
-                self.__deny("write_failed")
-            if await self.__read_native(node_id) is not None:
-                self.__deny("write_mismatch")
-            self.__deleted.add(node_id)
-            self.__permissions.pop(node_id, None)
-            self.__written.pop(node_id, None)
-            self.__superseded.pop(node_id, None)
-            self.__heads.pop(node_id, None)
-            return True
@@ -144,6 +144,10 @@ class HyReconcileGuard:
        self.failures: list[str] = []
        self.bindings: HyWriteBindings | None = None
        self.imported_contents: frozenset[str] | None = None
+        self.mutation_target_ids: set[str] = set()
+
+    def begin_write(self) -> None:
+        self.mutation_target_ids.clear()

    def __getattr__(self, name: str) -> Any:
        return getattr(self.reconciler, name)
@@ -210,6 +214,15 @@ class HyReconcileGuard:
        for operation in result.ops:
            raw_name = getattr(operation, "op", "")
            name = str(getattr(raw_name, "value", raw_name)).upper()
+            if name in {"UPDATE", "SUPERSEDE", "DELETE", "SHADOW"}:
+                target = getattr(operation, "memory_id", None)
+                if isinstance(target, str) and target:
+                    self.mutation_target_ids.add(target)
+                self.mutation_target_ids.update(
+                    key
+                    for key in getattr(operation, "memory_ids", ())
+                    if isinstance(key, str) and key
+                )
            if name == "ADD":
                operation_counts["ADD"] += 1
            elif name in {"UPDATE", "MERGE", "NOOP"}:
@@ -232,20 +245,6 @@ class HyReconcileGuard:
                "result_count": len(result.ops),
            },
        )
-        timer = self.step_timer
-        try:
-            with (
-                timer.step("write_verification", items=len(result.ops))
-                if timer is not None
-                else nullcontext()
-            ):
-                await self._check_targets(result, kwargs)
-        except Exception as error:
-            code = str(error)
-            if not code.startswith("hy_memory_write_verification_") or len(code) > 100:
-                code = "hy_memory_write_verification_failed"
-            self.failures.append(code)
-            raise ValueError(code) from error
        return result

    async def _check_targets(self, result: Any, kwargs: dict[str, Any]) -> None:
@@ -14,7 +14,7 @@ from pathlib import Path
 from typing import Literal, cast

 HY_MEMORY_DISTRIBUTION = "hy-memory"
-HY_MEMORY_VERSION = "2.0.1"
+HY_MEMORY_VERSION = "2.0.4"
 HY_MEMORY_NAMESPACE = "governed-v2"
 DiagnosticLogLevel = Literal["error", "warning", "info", "debug"]

@@ -31,11 +31,12 @@ _COLLECTION_LOGGERS = (
    "httpx",
    "httpcore",
 )
+_NOISY_LIBRARY_LOGGERS = ("httpx", "httpcore", "watchfiles", "jieba")

 # One Memory deliberately stays on ultra because its existing, user-visible
 # profile/graph flow calls digest(), which the pinned public SDK exposes only
 # for ultra clients. Reader and non-reranker behavior remain pinned for
-# compatibility with public SDK 2.0.1.
+# compatibility with public SDK 2.0.4.
 HY_MEMORY_MODE = "ultra"
 # The SDK only recognises `legacy` and `hybrid` (see
 # hy_memory/pipelines/_retrieval/config.py: `ALL_READERS = (READER_LEGACY,
@@ -139,6 +140,18 @@ def apply_diagnostic_log_level(level: DiagnosticLogLevel) -> None:
    for name in tuple(logging.Logger.manager.loggerDict):
        if any(name == root or name.startswith(f"{root}.") for root in _COLLECTION_LOGGERS):
            logging.getLogger(name).setLevel(selected)
+    # Successful HTTP exchanges, file-watch notifications and tokenizer
+    # initialization are useful only while actively debugging. At INFO they
+    # dominated core.log and the in-app panel without describing product work.
+    noisy_level = selected if selected == logging.DEBUG else max(selected, logging.WARNING)
+    for root in _NOISY_LIBRARY_LOGGERS:
+        logging.getLogger(root).setLevel(noisy_level)
+    for name in tuple(logging.Logger.manager.loggerDict):
+        if any(
+            name == root or name.startswith(f"{root}.")
+            for root in _NOISY_LIBRARY_LOGGERS
+        ):
+            logging.getLogger(name).setLevel(noisy_level)

    os.environ["MEMORY_LOG_LEVEL"] = logging.getLevelName(selected)
    os.environ["MEMORY_PIPELINE_TRACE_ENABLED"] = "true" if level == "debug" else "false"
@@ -7,7 +7,7 @@ from typing import Any, Literal

 from pydantic import BaseModel, ConfigDict

-from one_memory.imports.models import EvidenceContentKind, EvidenceRecord, WrittenBy
+from one_memory.imports.models import EvidenceContentKind, EvidenceRecord
 from one_memory.input_sanitization import sanitize_model_input


@@ -61,17 +61,7 @@ class HyWriteBindings:
                    seed.workspace,
                )
                or record.content_kind is not EvidenceContentKind.RAW_CONVERSATION
-                or (record.role, record.written_by)
-                not in {
-                    # D98: the same change as the turn assembly. A sub agent's
-                    # question is a `user` message marked written by the
-                    # agent; excluding it here would make the source context
-                    # disagree with what was actually sent.
-                    ("user", WrittenBy.PERSON),
-                    ("user", WrittenBy.AGENT),
-                    ("assistant", WrittenBy.AGENT),
-                    ("tool", WrittenBy.TOOL),
-                }
+                or record.role not in {"user", "assistant", "tool"}
            ):
                raise ValueError("hy_memory_write_verification_context_scope")
        self.all_sources: list[dict[str, Any]] = [