docs: announce Hy-Memory-30B-A3B FP8 and GGUF quantized models

Co-authored-by: Cursor <cursoragent@cursor.com>

CHANGELOG.md

@@ -14,6 +14,25 @@ Hy Memory and MemHub are versioned and released independently.
 
 ## Unreleased
 
+- Fixed governed-memory revisions and repeated evidence links failing when the
+  visibility dirty-marker trigger encounters the same memory twice. Startup
+  now replaces the affected legacy trigger without changing schema versions
+  or governed and visibility data.
+- Serialized each SDK SQLite store's shared connection lifecycle so concurrent
+  history/cache operations cannot race with `close()` and crash the process.
+- Scoped Coding reconciliation targets to the current HY authority,
+  conversation, and workspace, preventing near-duplicate sessions from
+  mutating an experience or working fact owned by another evidence context.
+- Kept post-write authority verification failures available for read-only
+  reconciliation while preventing automatic or generic confirmed retries from
+  invoking the model again. Automatic reconciliation skips those sticky rows
+  before applying its batch limit, and linked durable jobs exclusively own
+  retry scheduling for their memory tasks.
+- Clarified that `add(metadata=...)` carries legacy SDK processing options,
+  while persistent caller business context belongs in `meta`.
+- Added an opt-in `get(..., include_raw=True)` path for reading `L1_RAW`
+  memories, including memories written in lite mode. The default remains
+  hidden for backward compatibility.
 - 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.

README.md

@@ -32,7 +32,9 @@ English | [中文](README.zh.md)
 
 ## Latest News
 
-* *<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** quantized **FP8** (**[Hy-Memory-30B-A3B-FP8](https://huggingface.co/tencent/Hy-Memory-30B-A3B-FP8)**) and **GGUF** (**[Hy-Memory-30B-A3B-GGUF](https://huggingface.co/tencent/Hy-Memory-30B-A3B-GGUF)**) versions of **Hy-Memory-30B-A3B** on Hugging Face!
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 📦 **Released [MemHub v1.0.4](https://github.com/Tencent-Hunyuan/Hy-Memory/releases/tag/v1.0.4)** and **[memhub-cli 1.0.5](https://www.npmjs.com/package/memhub-cli/v/1.0.5)**, with a signed macOS desktop app and verified macOS/Linux Runtimes!
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **Released** **Hy Memory 2.0 (`hy-memory==2.0.4`)** on [PyPI](https://pypi.org/project/hy-memory/2.0.4/) 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)**!

README.zh.md

@@ -32,7 +32,9 @@
 
 ## 最新动态
 
-* *<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** 的 **FP8**（**[Hy-Memory-30B-A3B-FP8](https://huggingface.co/tencent/Hy-Memory-30B-A3B-FP8)**）与 **GGUF**（**[Hy-Memory-30B-A3B-GGUF](https://huggingface.co/tencent/Hy-Memory-30B-A3B-GGUF)**）量化版本模型！
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 📦 **正式发布 [MemHub v1.0.4](https://github.com/Tencent-Hunyuan/Hy-Memory/releases/tag/v1.0.4)** 与 **[memhub-cli 1.0.5](https://www.npmjs.com/package/memhub-cli/v/1.0.5)**，提供已签名的 macOS 桌面应用及经过验证的 macOS/Linux Runtime！
+* *<mark>[new]</mark>* 🔥🔥🔥 **[2026-10]**: 🚀 **正式发布** **Hy Memory 2.0 (`hy-memory==2.0.4`)**（已上线 [PyPI](https://pypi.org/project/hy-memory/2.0.4/)），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！

docs/usage.md

@@ -106,6 +106,15 @@ result = client.add(
 )
 ```
 
+`metadata` 与 `meta` 不是同一字段：
+
+- `metadata` 是兼容旧调用方保留的 SDK 处理选项，会进入内部
+  `WriteRequest.extra`，不保证持久化或随查询结果返回。
+- `meta` 是调用方的业务上下文，会持久化并在 `search()`、`get()` 等结果中返回。
+  同一记忆被多次合并时，各次 `meta` 会按写入顺序保存在列表中。
+
+需要以后读回的数据必须放在 `meta`，不要放在 `metadata`。
+
 `scene` 的公开取值：
 
 - `None`：普通文本直接走 chat；messages 含 tool 时执行快速场景判断。
@@ -263,6 +272,8 @@ catalog 也可以在没有基础 L0 时形成只读的合成画像。
 
 ```python
 memory = client.get("memory_id")
+# lite 模式写入的是 L1_RAW；按 ID 读取时必须显式选择返回原文。
+raw_memory = client.get("memory_id", include_raw=True)
 result = client.update("memory_id", "更新后的内容")
 result = client.delete("memory_id")
 result = client.delete_all(user_id="user_001")
@@ -282,6 +293,10 @@ result = client.clone_user("user_001", "user_002", agent_id=None)
 对应异步方法为 `async_get()`、`async_update()`、`async_delete()`、
 `async_delete_all()`、`async_list_memories()`、`async_clone_user()`。
 
+`get()` / `async_get()` 默认继续隐藏 `L1_RAW`，因此不会改变现有调用方的原文访问
+边界。只有显式传入 `include_raw=True` 才返回该层；这也是 lite 模式按 ID 读取
+写入内容及其 `meta` 的方式。MemHub 的 HTTP 读取接口不开放此参数。
+
 普通 `list_memories()` 返回 `vdb` 桶，并在 Graph 有结果时附带 `graph`；传入
 `workspace_id` 时返回生产力记忆的 `scene="productivity"` 和 `experiences`。
 

hy_memory/client.py

@@ -1597,7 +1597,9 @@ class HyMemoryClient:
             session_id:   Session ID（三级隔离 key）
             source:       调用来源，例如 openclaw / cursor / codex；与 agent_id 正交。
                           写入前执行 strip().lower()，默认 "default"。
-            metadata:     自定义元数据
+            metadata:     旧版 SDK 处理选项，会并入内部 WriteRequest.extra。
+                          不保证持久化或随查询返回；调用方需要保存并读回的业务字段
+                          请使用 meta。
             memory_at:    记忆时间戳（不传则用当前时间），用于导入历史记忆
             scene:        显式指定写入路由，跳过 LLM 自动判定。
                           - None（默认）：含 tool 时 LLM 判 chat/working/coding；
@@ -1623,8 +1625,8 @@ class HyMemoryClient:
             branch:       Coding memory 路径专用。分支名。同 workspace_id。chat 路径忽略。
             meta:         调用方自定义业务 meta，随本次产出的记忆存储、查询时原样返回。
                           不 embed、不参与召回、不作过滤条件。
-                          与 metadata 不同：metadata 会并入 extra（含 SDK 内部控制键），
+                          与 metadata 不同：metadata 是旧版内部处理选项，
-                          meta 独立存放，只做透传。
+                          meta 独立持久化，只做业务透传。
                           语义为「该次 write 时刻的上下文」，故存为 list——同一节点被
                           多次写入合并时会追加而非覆盖，详见 get() 返回的 meta 字段。
 
@@ -1791,17 +1793,26 @@ class HyMemoryClient:
             query_embedding=query_embedding,
         ))
 
-    def get(self, memory_id: str) -> Optional[Dict[str, Any]]:
+    def get(
+        self,
+        memory_id: str,
+        *,
+        include_raw: bool = False,
+    ) -> Optional[Dict[str, Any]]:
         """
         获取单条记忆。
 
         Args:
             memory_id: 记忆 ID
+            include_raw: 是否允许返回 L1_RAW 原始记忆。默认 False，保持原有隐藏行为；
+                         lite 模式的记忆均为 L1_RAW，需显式设为 True 才能按 ID 读取。
 
         Returns:
             {"memory_id", "content", "layer", "request_id", "elapsed_ms"} 或 None（不存在时）
         """
-        return self._loop_thread.run(self.async_get(memory_id))
+        return self._loop_thread.run(
+            self.async_get(memory_id, include_raw=include_raw)
+        )
 
     def update(
         self,
@@ -3650,12 +3661,21 @@ class HyMemoryClient:
     # ================================================================
 
     @_ensure_internal_loop
-    async def async_get(self, memory_id: str) -> Optional[Dict[str, Any]]:
+    async def async_get(
+        self,
+        memory_id: str,
+        *,
+        include_raw: bool = False,
+    ) -> Optional[Dict[str, Any]]:
         """
         获取单条记忆（异步）。
 
         VDB + Graph 双路并行查询，哪个有结果返回哪个。
 
+        Args:
+            memory_id: 记忆 ID
+            include_raw: 是否允许返回 L1_RAW 原始记忆，默认 False。
+
         Returns:
             {"memory_id", "content", "layer", "source", "request_id", "elapsed_ms"} 或 None
         """
@@ -3683,17 +3703,24 @@ class HyMemoryClient:
 
         vdb_node, graph_node = await _asyncio_get.gather(_get_vdb(), _get_graph())
 
-        # L1_RAW 不对外可见：按 ID 直取走主键查询、绕过向量库的 layer filter，
+        # L1_RAW 默认不对外可见：按 ID 直取走主键查询、绕过向量库的 layer
-        # 故在此统一挡掉。async_get 是 GET /v1/memories/{id} 与
+        # filter，故在此统一挡掉。SDK 调用方可通过 include_raw=True 显式读取；
-        # /v1/memories/batch 两个对外接口的共同必经处，而内部写链路不经过它
+        # HTTP 接口不传该参数，继续保持隐藏。内部写链路不经过此方法。
-        # （writer 写后校验直接调 vector_store.get_by_id），挡在此处不影响内部。
         #
         # 置于 gather 之后、埋点之前：命中 raw 仍照常落 GET 埋点（保留可观测性，
         # 便于发现有调用方拿 raw id 试探），只是不把内容返回给调用方。
         # 对调用方的表现与"该 id 不存在"完全一致，不额外暴露"存在但被挡"的信息。
-        if vdb_node is not None and vdb_node.layer == MemoryLayer.L1_RAW:
+        if (
+            not include_raw
+            and vdb_node is not None
+            and vdb_node.layer == MemoryLayer.L1_RAW
+        ):
             vdb_node = None
-        if graph_node is not None and getattr(graph_node, "layer", None) == MemoryLayer.L1_RAW:
+        if (
+            not include_raw
+            and graph_node is not None
+            and getattr(graph_node, "layer", None) == MemoryLayer.L1_RAW
+        ):
             graph_node = None
 
         elapsed = round((time.perf_counter() - t0) * 1000, 2)

hy_memory/coding/fact_reconciler.py

@@ -466,7 +466,13 @@ class WorkingFactReconciler:
             return [], []
 
         candidates = await self._retrieve_candidates(
-            drafts, user_id=user_id, source=source,
+            drafts,
+            user_id=user_id,
+            agent_id=agent_id,
+            session_id=session_id,
+            workspace_id=workspace_id,
+            branch=branch,
+            source=source,
         )
 
         try:
@@ -499,7 +505,15 @@ class WorkingFactReconciler:
     # ------------------------------------------------------------
 
     async def _retrieve_candidates(
-        self, drafts: List[WorkingFactDraft], *, user_id: str, source: str
+        self,
+        drafts: List[WorkingFactDraft],
+        *,
+        user_id: str,
+        agent_id: str,
+        session_id: Optional[str],
+        workspace_id: Optional[str],
+        branch: Optional[str],
+        source: str,
     ) -> List[List[WorkingFact]]:
         """每条 draft 用 statement 召回同 user 的 top-N active 事实。
 
@@ -512,6 +526,10 @@ class WorkingFactReconciler:
         关掉时整份召回原样交给 LLM。适用于 subject 不是稳定实体名的数据集（自由文本、
         位置标签），那里精确字符串相等会把同实体判成不同主体，反而拦掉真实更新。
         """
+        # Empty/legacy session metadata remains readable, but cannot prove
+        # conversation authority for a destructive SUPERSEDE.
+        if not session_id:
+            return [[] for _ in drafts]
         try:
             embeddings = await self.embed_service.embed_batch(
                 [d.statement for d in drafts]
@@ -525,6 +543,10 @@ class WorkingFactReconciler:
             try:
                 hits = await self.store.search_facts(
                     emb, user_id=user_id, top=self.candidates_per_draft,
+                    agent_id=agent_id,
+                    session_id=session_id,
+                    workspace_id=workspace_id,
+                    branch=branch,
                     # 写入侧同样接词面通道：候选选错了 LLM 就没有机会判对。这里不传
                     # min_score——写入侧要的是"有哪些可能的前身"，让 LLM 看更宽的名单，
                     # 阈值过滤是读取侧的事。
@@ -794,6 +816,15 @@ class WorkingFactReconciler:
                 old = await self.store.get_fact(old_id, user_id=user_id)
                 if old is None:
                     continue
+                if (
+                    old.agent_id != agent_id
+                    or old.session_id != session_id
+                    or old.workspace_id != workspace_id
+                ):
+                    logger.warning(
+                        f"[fact-reconcile] refused cross-authority SUPERSEDE: {old.fact_id}"
+                    )
+                    continue
                 # 闸门在候选阶段已经拦过一道，这里再拦一次：候选是按 statement 召回的，
                 # 而 LLM 只能覆盖候选名单里的 id，理论上到不了这儿。真到了就说明前面
                 # 哪一环漏了，宁可让这次覆盖不生效——多留一条旧事实是可恢复的，删掉

hy_memory/coding/reconciler.py

@@ -240,7 +240,7 @@ class CodingMemoryReconciler:
         )
 
         # 3. 校验 + 执行
-        executed = await self._execute(ops, drafts)
+        executed = await self._execute(ops, drafts, candidates_per_draft)
         logger.info(
             f"[coding-reconcile] drafts={len(drafts)} ops={len(executed)} "
             + ", ".join(f"{a}={sum(1 for o in executed if o.action == a)}"
@@ -263,6 +263,12 @@ class CodingMemoryReconciler:
         """
         result: List[List[CodingMemory]] = []
         for d in drafts:
+            # A coding mutation must be attributable to one conversation.
+            # Empty/legacy session metadata stays readable through ordinary
+            # search, but it cannot authorize UPDATE/DELETE.
+            if not d.session_id:
+                result.append([])
+                continue
             try:
                 # task 为主入口，search_keys 为附加入口（去重、去空）
                 queries = [d.task] + [k for k in (d.search_keys or []) if k and k != d.task]
@@ -274,6 +280,8 @@ class CodingMemoryReconciler:
                     hits = await self.store.search_by_query_embedding(
                         emb,
                         user_id=d.user_id,
+                        agent_id=d.agent_id,
+                        session_id=d.session_id,
                         workspace_id=d.workspace_id,
                         branch=d.branch,
                         top=self.candidates_per_draft * 4,
@@ -415,6 +423,7 @@ class CodingMemoryReconciler:
         self,
         ops: List[ReconcileOp],
         drafts: List[CodingMemoryDraft],
+        candidates_per_draft: List[List[CodingMemory]],
     ) -> List[ReconcileOp]:
         """
         校验 + 执行 ops。返回成功执行的 ops（filter 掉无效 op）。
@@ -454,6 +463,11 @@ class CodingMemoryReconciler:
                 if not op.target_experience_id:
                     logger.warning(f"[coding-reconcile] drop UPDATE missing target: {op.to_dict()}")
                     continue
+                if not self._target_was_offered(op, candidates_per_draft):
+                    logger.warning(
+                        f"[coding-reconcile] drop UPDATE outside authority scope: {op.to_dict()}"
+                    )
+                    continue
                 if op.draft_idx in consumed:
                     logger.warning(f"[coding-reconcile] drop UPDATE on consumed draft: {op.to_dict()}")
                     continue
@@ -476,6 +490,11 @@ class CodingMemoryReconciler:
                 if not op.target_experience_id:
                     logger.warning(f"[coding-reconcile] drop DELETE missing target: {op.to_dict()}")
                     continue
+                if not self._target_was_offered(op, candidates_per_draft):
+                    logger.warning(
+                        f"[coding-reconcile] drop DELETE outside authority scope: {op.to_dict()}"
+                    )
+                    continue
                 # DELETE 需要知道 user_id；通过 draft_idx（如有）或第一个 draft 推断
                 user_id = self._infer_user_id(op, drafts)
                 if not user_id:
@@ -512,6 +531,19 @@ class CodingMemoryReconciler:
 
         return executed
 
+    @staticmethod
+    def _target_was_offered(
+        op: ReconcileOp,
+        candidates_per_draft: List[List[CodingMemory]],
+    ) -> bool:
+        """Only mutate a target frozen into this draft's scoped candidate set."""
+        if op.draft_idx is None or not (0 <= op.draft_idx < len(candidates_per_draft)):
+            return False
+        return any(
+            candidate.experience_id == op.target_experience_id
+            for candidate in candidates_per_draft[op.draft_idx]
+        )
+
     @staticmethod
     def _infer_user_id(
         op: ReconcileOp, drafts: List[CodingMemoryDraft]

hy_memory/coding/store.py

@@ -529,10 +529,20 @@ class CodingMemoryStore:
         """覆盖式更新一条 coding memory。"""
         # 拿旧 keys 长度，先删旧 VDB rows
         prev = await self.get_by_id(memory.experience_id, user_id=memory.user_id)
-        prev_key_count = (1 + len(prev.search_keys)) if prev else 0
+        if prev is None:
+            raise ValueError("coding_memory_update_target_missing")
+        if (
+            not prev.session_id
+            or not memory.session_id
+            or prev.agent_id != memory.agent_id
+            or prev.session_id != memory.session_id
+            or prev.workspace_id != memory.workspace_id
+        ):
+            raise ValueError("coding_memory_update_authority_mismatch")
+        prev_key_count = 1 + len(prev.search_keys)
 
         memory.updated_at = datetime.utcnow()
-        if memory.created_at is None and prev is not None:
+        if memory.created_at is None:
             memory.created_at = prev.created_at
         await _run_in_pool(self._upsert_sql_sync, memory)
         # 删旧 VDB
@@ -671,6 +681,8 @@ class CodingMemoryStore:
         query_embedding: List[float],
         user_id: str,
         *,
+        agent_id: Optional[str] = None,
+        session_id: Optional[str] = None,
         workspace_id: Optional[str] = None,
         branch: Optional[str] = None,
         top: int = 40,
@@ -688,21 +700,28 @@ class CodingMemoryStore:
                 },
                 ...
             ]
-        workspace_id / branch 是作用域过滤，规则见 _scope_where：不传则纯语义召回
+        agent_id / session_id 是写授权作用域，传入时精确匹配；workspace_id / branch
-        （行为与加过滤前完全一致）；传了则只保留 SQLite 里落在作用域内的经验，并按
+        的规则见 _scope_where。不传任何作用域则维持纯语义召回；传了则只保留 SQLite
-        作用域的选择率放大向量取数，避免"先取 top-N 再过滤"把召回吃掉。
+        里落在作用域内的经验，并按选择率放大向量取数，避免"先取 top-N 再过滤"把
+        召回吃掉。
         """
         if self._vec is None:
             return []
 
         allowed: Optional[Set[str]] = None
-        if workspace_id:
+        if agent_id or session_id or workspace_id:
             allowed, total = await _run_in_pool(
-                self._scoped_experience_ids_sync, user_id, workspace_id, branch
+                self._authority_scoped_experience_ids_sync,
+                user_id,
+                agent_id,
+                session_id,
+                workspace_id,
+                branch,
             )
             if not allowed:
                 logger.info(
                     f"[coding-store] no experience in scope: "
+                    f"agent_id={agent_id!r} session_id={session_id!r} "
                     f"workspace_id={workspace_id!r} branch={branch!r} "
                     f"(user has {total}); skip vdb search"
                 )
@@ -778,6 +797,8 @@ class CodingMemoryStore:
         query_embedding: List[float],
         user_id: str,
         *,
+        agent_id: Optional[str] = None,
+        session_id: Optional[str] = None,
         workspace_id: Optional[str] = None,
         branch: Optional[str] = None,
         top: int = 5,
@@ -791,8 +812,9 @@ class CodingMemoryStore:
         SUPERSEDE 时向量行已被删掉，理论上召不到 superseded；这里再按 SQLite 的
         status 过一道，避免两边不一致时把过期结论放出去。
 
-        workspace_id / branch 是作用域过滤，规则见 _scope_where，语义与经验层一致：
+        agent_id / session_id 是写授权作用域，传入时精确匹配；workspace_id / branch
-        不传不过滤，传了按 SQLite 里的值筛，并按选择率放大向量取数。
+        的规则见 _scope_where。不传任何作用域则不过滤，传了按 SQLite 里的值筛，并
+        按选择率放大向量取数。
 
         向量库按 top 的 4 倍过采样：向量行和 SQLite 行不是一个事务里写的，中间挂掉
         就会留下"向量还在、SQLite 没了"的孤儿，软删时删向量失败也会留下 superseded
@@ -809,13 +831,19 @@ class CodingMemoryStore:
             return []
 
         fetch = top * 4
-        if workspace_id:
+        if agent_id or session_id or workspace_id:
             scoped, total = await _run_in_pool(
-                self._count_active_facts_sync, user_id, workspace_id, branch
+                self._count_active_facts_sync,
+                user_id,
+                agent_id,
+                session_id,
+                workspace_id,
+                branch,
             )
             if not scoped:
                 logger.info(
                     f"[coding-store] no active fact in scope: "
+                    f"agent_id={agent_id!r} session_id={session_id!r} "
                     f"workspace_id={workspace_id!r} branch={branch!r} "
                     f"(user has {total}); skip fact vdb search"
                 )
@@ -852,6 +880,8 @@ class CodingMemoryStore:
         alive = [
             (f, scores[f.fact_id]) for f in facts
             if f.status == "active"
+            and (not agent_id or f.agent_id == agent_id)
+            and (not session_id or f.session_id == session_id)
             and _scope_matches(f.workspace_id, f.branch, workspace_id, branch)
         ]
         kept = [(f, s) for f, s in alive if s >= min_score]
@@ -1050,14 +1080,25 @@ class CodingMemoryStore:
             by_id = {r["experience_id"]: _row_to_memory(r) for r in rows}
             return [by_id[eid] for eid in experience_ids if eid in by_id]
 
-    def _scoped_experience_ids_sync(
+    def _authority_scoped_experience_ids_sync(
-        self, user_id: str, workspace_id: str, branch: Optional[str]
+        self,
+        user_id: str,
+        agent_id: Optional[str],
+        session_id: Optional[str],
+        workspace_id: Optional[str],
+        branch: Optional[str],
     ) -> Tuple[Set[str], int]:
-        """返回 (落在作用域内的 experience_id 集合, 该 user 的经验总数)。
+        """返回 (落在授权作用域内的 experience_id 集合, 该 user 的经验总数)。
 
         一次查询同时拿到过滤结果和选择率分母：前者供向量命中做交集，后者定放大倍数。
         """
         scope_sql, scope_params = _scope_where(workspace_id, branch)
+        if agent_id:
+            scope_sql = f"agent_id = ? AND ({scope_sql})"
+            scope_params = [agent_id, *scope_params]
+        if session_id:
+            scope_sql = f"session_id = ? AND ({scope_sql})"
+            scope_params = [session_id, *scope_params]
         with self._lock:
             assert self._conn is not None
             cur = self._conn.execute(
@@ -1069,13 +1110,24 @@ class CodingMemoryStore:
         return {r["experience_id"] for r in rows if r["in_scope"]}, len(rows)
 
     def _count_active_facts_sync(
-        self, user_id: str, workspace_id: str, branch: Optional[str]
+        self,
+        user_id: str,
+        agent_id: Optional[str],
+        session_id: Optional[str],
+        workspace_id: Optional[str],
+        branch: Optional[str],
     ) -> Tuple[int, int]:
         """返回 (作用域内 active 事实数, 该 user 的 active 事实总数)。
 
         事实层的过滤在 get_facts 拿到对象之后用 _scope_matches 做，这里只要选择率。
         """
         scope_sql, scope_params = _scope_where(workspace_id, branch)
+        if agent_id:
+            scope_sql = f"agent_id = ? AND ({scope_sql})"
+            scope_params = [agent_id, *scope_params]
+        if session_id:
+            scope_sql = f"session_id = ? AND ({scope_sql})"
+            scope_params = [session_id, *scope_params]
         with self._lock:
             assert self._conn is not None
             row = self._conn.execute(

hy_memory/data/cache_sqlite.py

@@ -7,7 +7,7 @@ Agent Memory V2 - SqliteCache
 设计模式跟随 history_store.py:
 - 标准库 sqlite3 + _run_in_sqlite_pool()
 - WAL 模式 + check_same_thread=False
-- threading.Lock 保护写操作
+- 每实例 threading.RLock 保护共享连接的全部操作与关闭
 
 三张表:
 - memory_operations / pipeline_logs / system_metrics: 审计与观测
@@ -122,7 +122,10 @@ class SqliteCache(CacheBase):
     def __init__(self, config: MemoryConfig):
         self.config = config
         self._conn: Optional[sqlite3.Connection] = None
-        self._lock = threading.Lock()
+        # One connection is shared by the executor workers. Use the same lock
+        # for reads, writes, initialization and close so close() cannot free
+        # the connection while another worker is inside sqlite3.
+        self._lock = threading.RLock()
 
         # 读取 db_path
         cache_cfg = getattr(config, "cache", None)
@@ -144,35 +147,37 @@ class SqliteCache(CacheBase):
         logger.debug(f"SqliteCache initialized: {self._db_path}")
 
     def _init_sync(self) -> None:
-        db_dir = os.path.dirname(self._db_path)
+        with self._lock:
-        if db_dir:
+            db_dir = os.path.dirname(self._db_path)
-            os.makedirs(db_dir, exist_ok=True)
+            if db_dir:
-        self._conn = sqlite3.connect(self._db_path, check_same_thread=False)
+                os.makedirs(db_dir, exist_ok=True)
-        self._conn.row_factory = sqlite3.Row
+            self._conn = sqlite3.connect(self._db_path, check_same_thread=False)
-        self._conn.execute("PRAGMA journal_mode=WAL;")
+            self._conn.row_factory = sqlite3.Row
-        self._conn.executescript(_CREATE_TABLES_SQL)
+            self._conn.execute("PRAGMA journal_mode=WAL;")
-        self._migrate_pipeline_logs_columns()
+            self._conn.executescript(_CREATE_TABLES_SQL)
-        self._conn.commit()
+            self._migrate_pipeline_logs_columns()
+            self._conn.commit()
 
     def _migrate_pipeline_logs_columns(self) -> None:
         """给旧库的 pipeline_logs_v2 补 namespace/session/event time 列。"""
-        try:
+        with self._lock:
-            cols = {
+            try:
-                row["name"]
+                cols = {
-                for row in self._conn.execute("PRAGMA table_info(pipeline_logs_v2)")
+                    row["name"]
-            }
+                    for row in self._conn.execute("PRAGMA table_info(pipeline_logs_v2)")
-            _text_cols = ("namespace", "session_id")
+                }
-            for col in _text_cols:
+                _text_cols = ("namespace", "session_id")
-                if col not in cols:
+                for col in _text_cols:
+                    if col not in cols:
+                        self._conn.execute(
+                            f"ALTER TABLE pipeline_logs_v2 ADD COLUMN {col} TEXT NOT NULL DEFAULT ''"
+                        )
+                if "event_at_ms" not in cols:
                     self._conn.execute(
-                        f"ALTER TABLE pipeline_logs_v2 ADD COLUMN {col} TEXT NOT NULL DEFAULT ''"
+                        "ALTER TABLE pipeline_logs_v2 ADD COLUMN event_at_ms INTEGER NOT NULL DEFAULT 0"
                     )
-            if "event_at_ms" not in cols:
+            except Exception as e:
-                self._conn.execute(
+                logger.debug(f"[sqlite] pipeline_logs_v2 column migration skipped: {e}")
-                    "ALTER TABLE pipeline_logs_v2 ADD COLUMN event_at_ms INTEGER NOT NULL DEFAULT 0"
-                )
-        except Exception as e:
-            logger.debug(f"[sqlite] pipeline_logs_v2 column migration skipped: {e}")
 
     async def close(self) -> None:
         if self._conn is not None:
@@ -180,9 +185,10 @@ class SqliteCache(CacheBase):
         logger.info("SqliteCache closed")
 
     def _close_sync(self) -> None:
-        if self._conn:
+        with self._lock:
-            self._conn.close()
+            if self._conn:
-            self._conn = None
+                self._conn.close()
+                self._conn = None
 
     # ================================================================
     # 统计
@@ -255,39 +261,40 @@ class SqliteCache(CacheBase):
         """
         try:
             def _query():
-                conditions = []
+                with self._lock:
-                params = []
+                    conditions = []
-                if request_id:
+                    params = []
-                    conditions.append("request_id = ?")
+                    if request_id:
-                    params.append(request_id)
+                        conditions.append("request_id = ?")
-                if memory_id:
+                        params.append(request_id)
-                    conditions.append("memory_id = ?")
+                    if memory_id:
-                    params.append(memory_id)
+                        conditions.append("memory_id = ?")
-                if user_id:
+                        params.append(memory_id)
-                    conditions.append("user_id = ?")
+                    if user_id:
-                    params.append(user_id)
+                        conditions.append("user_id = ?")
+                        params.append(user_id)
 
-                if not conditions:
+                    if not conditions:
-                    return []
+                        return []
 
-                where = " AND ".join(conditions)
+                    where = " AND ".join(conditions)
-                rows = self._conn.execute(
+                    rows = self._conn.execute(
-                    f"SELECT * FROM memory_operations_v2 WHERE {where} "
+                        f"SELECT * FROM memory_operations_v2 WHERE {where} "
-                    f"ORDER BY id DESC LIMIT ?",
+                        f"ORDER BY id DESC LIMIT ?",
-                    params + [limit],
+                        params + [limit],
-                ).fetchall()
+                    ).fetchall()
-                result = []
+                    result = []
-                for r in rows:
+                    for r in rows:
-                    row = dict(r)
+                        row = dict(r)
-                    # 反序列化 supersedes JSON 字符串 → list
+                        # 反序列化 supersedes JSON 字符串 → list
-                    import json as _json
+                        import json as _json
-                    raw_sup = row.get("supersedes", "")
+                        raw_sup = row.get("supersedes", "")
-                    try:
+                        try:
-                        row["supersedes"] = _json.loads(raw_sup) if raw_sup else []
+                            row["supersedes"] = _json.loads(raw_sup) if raw_sup else []
-                    except Exception:
+                        except Exception:
-                        row["supersedes"] = []
+                            row["supersedes"] = []
-                    result.append(row)
+                        result.append(row)
-                return result
+                    return result
 
             return await _run_in_sqlite_pool(_query)
         except Exception as e:
@@ -369,37 +376,38 @@ class SqliteCache(CacheBase):
         """
         try:
             def _query():
-                conditions = []
+                with self._lock:
-                params = []
+                    conditions = []
-                if request_id:
+                    params = []
-                    conditions.append("request_id = ?")
+                    if request_id:
-                    params.append(request_id)
+                        conditions.append("request_id = ?")
-                if user_id:
+                        params.append(request_id)
-                    conditions.append("user_id = ?")
+                    if user_id:
-                    params.append(user_id)
+                        conditions.append("user_id = ?")
-                if step:
+                        params.append(user_id)
-                    conditions.append("step = ?")
+                    if step:
-                    params.append(step)
+                        conditions.append("step = ?")
+                        params.append(step)
 
-                if not conditions:
+                    if not conditions:
-                    return []
+                        return []
 
-                where = " AND ".join(conditions)
+                    where = " AND ".join(conditions)
-                rows = self._conn.execute(
+                    rows = self._conn.execute(
-                    f"SELECT * FROM pipeline_logs_v2 WHERE {where} "
+                        f"SELECT * FROM pipeline_logs_v2 WHERE {where} "
-                    f"ORDER BY id ASC LIMIT ?",
+                        f"ORDER BY id ASC LIMIT ?",
-                    params + [limit],
+                        params + [limit],
-                ).fetchall()
+                    ).fetchall()
-                out = [dict(r) for r in rows]
+                    out = [dict(r) for r in rows]
-                # 派生可读的 event_at（ISO ms）供查询接口直接用；event_at_ms 为 0 时留空
+                    # 派生可读的 event_at（ISO ms）供查询接口直接用；event_at_ms 为 0 时留空
-                from datetime import datetime as _dt
+                    from datetime import datetime as _dt
-                for _r in out:
+                    for _r in out:
-                    _ms = _r.get("event_at_ms") or 0
+                        _ms = _r.get("event_at_ms") or 0
-                    _r["event_at"] = (
+                        _r["event_at"] = (
-                        _dt.fromtimestamp(_ms / 1000.0).isoformat(timespec="milliseconds")
+                            _dt.fromtimestamp(_ms / 1000.0).isoformat(timespec="milliseconds")
-                        if _ms else ""
+                            if _ms else ""
-                    )
+                        )
-                return out
+                    return out
 
             return await _run_in_sqlite_pool(_query)
         except Exception as e:
@@ -414,26 +422,27 @@ class SqliteCache(CacheBase):
         """存储一个分钟桶的增量指标数据（UPSERT: 同一分钟多次写入会合并）"""
         try:
             def _store():
-                now = datetime.now().isoformat()
+                with self._lock:
-                # UPSERT: 如果同一分钟已存在，合并数据
+                    now = datetime.now().isoformat()
-                existing = self._conn.execute(
+                    # UPSERT: 如果同一分钟已存在，合并数据
-                    "SELECT data FROM system_metrics WHERE minute_ts = ?",
+                    existing = self._conn.execute(
-                    (minute_ts,),
+                        "SELECT data FROM system_metrics WHERE minute_ts = ?",
-                ).fetchone()
+                        (minute_ts,),
-                if existing:
+                    ).fetchone()
-                    import json as _json
+                    if existing:
-                    old_data = _json.loads(existing["data"])
+                        import json as _json
-                    merged = self._merge_metric_buckets(old_data, data)
+                        old_data = _json.loads(existing["data"])
-                    self._conn.execute(
+                        merged = self._merge_metric_buckets(old_data, data)
-                        "UPDATE system_metrics SET data = ?, created_at = ? WHERE minute_ts = ?",
+                        self._conn.execute(
-                        (_json.dumps(merged, default=str), now, minute_ts),
+                            "UPDATE system_metrics SET data = ?, created_at = ? WHERE minute_ts = ?",
-                    )
+                            (_json.dumps(merged, default=str), now, minute_ts),
-                else:
+                        )
-                    self._conn.execute(
+                    else:
-                        "INSERT INTO system_metrics (minute_ts, data, created_at) VALUES (?, ?, ?)",
+                        self._conn.execute(
-                        (minute_ts, json.dumps(data, default=str), now),
+                            "INSERT INTO system_metrics (minute_ts, data, created_at) VALUES (?, ?, ?)",
-                    )
+                            (minute_ts, json.dumps(data, default=str), now),
-                self._conn.commit()
+                        )
+                    self._conn.commit()
             await _run_in_sqlite_pool(_store)
         except Exception as e:
             logger.warning(f"store_metrics_minute failed: {e}")
@@ -442,17 +451,18 @@ class SqliteCache(CacheBase):
         """读取指定时间范围内的分钟指标数据"""
         try:
             def _load():
-                rows = self._conn.execute(
+                with self._lock:
-                    "SELECT data FROM system_metrics WHERE minute_ts >= ? AND minute_ts <= ? ORDER BY minute_ts",
+                    rows = self._conn.execute(
-                    (start_ts, end_ts),
+                        "SELECT data FROM system_metrics WHERE minute_ts >= ? AND minute_ts <= ? ORDER BY minute_ts",
-                ).fetchall()
+                        (start_ts, end_ts),
-                results = []
+                    ).fetchall()
-                for row in rows:
+                    results = []
-                    try:
+                    for row in rows:
-                        results.append(json.loads(row["data"]))
+                        try:
-                    except (json.JSONDecodeError, TypeError):
+                            results.append(json.loads(row["data"]))
-                        pass
+                        except (json.JSONDecodeError, TypeError):
-                return results
+                            pass
+                    return results
             return await _run_in_sqlite_pool(_load)
         except Exception as e:
             logger.warning(f"load_metrics_range failed: {e}")
@@ -462,11 +472,12 @@ class SqliteCache(CacheBase):
         """删除指定时间之前的 metrics 数据"""
         try:
             def _cleanup():
-                self._conn.execute(
+                with self._lock:
-                    "DELETE FROM system_metrics WHERE minute_ts < ?",
+                    self._conn.execute(
-                    (before_ts,),
+                        "DELETE FROM system_metrics WHERE minute_ts < ?",
-                )
+                        (before_ts,),
-                self._conn.commit()
+                    )
+                    self._conn.commit()
             await _run_in_sqlite_pool(_cleanup)
         except Exception as e:
             logger.warning(f"cleanup_old_metrics failed: {e}")

hy_memory/data/history_store.py

@@ -16,6 +16,7 @@ import json
 import logging
 import os
 import sqlite3
+import threading
 from datetime import datetime, timezone
 from typing import Dict, List, Any, Optional
 
@@ -89,6 +90,10 @@ class HistoryStore:
         """
         self._db_path = config.history.db_path
         self._conn: Optional[sqlite3.Connection] = None
+        # sqlite3.Connection.close() must never race with an execute() on the
+        # same connection. The lock also serializes operations on this shared
+        # connection; separate HistoryStore instances remain independent.
+        self._conn_lock = threading.RLock()
 
     async def initialize(self) -> None:
         """创建表和索引（如果不存在）"""
@@ -97,20 +102,21 @@ class HistoryStore:
 
     def _init_sync(self) -> None:
         """同步初始化：创建目录、连接数据库、建表建索引"""
-        # 确保目录存在
+        with self._conn_lock:
-        db_dir = os.path.dirname(self._db_path)
+            # 确保目录存在
-        if db_dir:
+            db_dir = os.path.dirname(self._db_path)
-            os.makedirs(db_dir, exist_ok=True)
+            if db_dir:
+                os.makedirs(db_dir, exist_ok=True)
 
-        self._conn = sqlite3.connect(self._db_path, check_same_thread=False)
+            self._conn = sqlite3.connect(self._db_path, check_same_thread=False)
-        self._conn.row_factory = sqlite3.Row
+            self._conn.row_factory = sqlite3.Row
-        # 开启 WAL 模式，提高并发读写性能
+            # 开启 WAL 模式，提高并发读写性能
-        self._conn.execute("PRAGMA journal_mode=WAL;")
+            self._conn.execute("PRAGMA journal_mode=WAL;")
 
-        self._conn.execute(_CREATE_TABLE_SQL)
+            self._conn.execute(_CREATE_TABLE_SQL)
-        for idx_sql in _CREATE_INDEXES_SQL:
+            for idx_sql in _CREATE_INDEXES_SQL:
-            self._conn.execute(idx_sql)
+                self._conn.execute(idx_sql)
-        self._conn.commit()
+            self._conn.commit()
 
     # ================================================================
     # 记录操作
@@ -218,13 +224,14 @@ class HistoryStore:
         return await _run_in_sqlite_pool(self._get_history_sync, memory_id)
 
     def _get_history_sync(self, memory_id: str) -> List[Dict[str, Any]]:
-        if self._conn is None:
+        with self._conn_lock:
-            return []
+            if self._conn is None:
-        cursor = self._conn.execute(
+                return []
-            "SELECT * FROM memory_history WHERE memory_id = ? ORDER BY created_at ASC",
+            cursor = self._conn.execute(
-            (memory_id,),
+                "SELECT * FROM memory_history WHERE memory_id = ? ORDER BY created_at ASC",
-        )
+                (memory_id,),
-        return [dict(row) for row in cursor.fetchall()]
+            )
+            return [dict(row) for row in cursor.fetchall()]
 
     async def get_recent(
         self,
@@ -244,14 +251,15 @@ class HistoryStore:
         return await _run_in_sqlite_pool(self._get_recent_sync, isolation_key, limit)
 
     def _get_recent_sync(self, isolation_key: str, limit: int) -> List[Dict[str, Any]]:
-        if self._conn is None:
+        with self._conn_lock:
-            return []
+            if self._conn is None:
-        cursor = self._conn.execute(
+                return []
-            "SELECT * FROM memory_history WHERE isolation_key = ? "
+            cursor = self._conn.execute(
-            "ORDER BY created_at DESC LIMIT ?",
+                "SELECT * FROM memory_history WHERE isolation_key = ? "
-            (isolation_key, limit),
+                "ORDER BY created_at DESC LIMIT ?",
-        )
+                (isolation_key, limit),
-        return [dict(row) for row in cursor.fetchall()]
+            )
+            return [dict(row) for row in cursor.fetchall()]
 
     async def count(self, isolation_key: str = "") -> int:
         """
@@ -266,16 +274,17 @@ class HistoryStore:
         return await _run_in_sqlite_pool(self._count_sync, isolation_key)
 
     def _count_sync(self, isolation_key: str) -> int:
-        if self._conn is None:
+        with self._conn_lock:
-            return 0
+            if self._conn is None:
-        if isolation_key:
+                return 0
-            cursor = self._conn.execute(
+            if isolation_key:
-                "SELECT COUNT(*) FROM memory_history WHERE isolation_key = ?",
+                cursor = self._conn.execute(
-                (isolation_key,),
+                    "SELECT COUNT(*) FROM memory_history WHERE isolation_key = ?",
-            )
+                    (isolation_key,),
-        else:
+                )
-            cursor = self._conn.execute("SELECT COUNT(*) FROM memory_history")
+            else:
-        return cursor.fetchone()[0]
+                cursor = self._conn.execute("SELECT COUNT(*) FROM memory_history")
+            return cursor.fetchone()[0]
 
     async def close(self) -> None:
         """关闭数据库连接"""
@@ -284,9 +293,10 @@ class HistoryStore:
             logger.info("[HistoryStore] closed")
 
     def _close_sync(self) -> None:
-        if self._conn:
+        with self._conn_lock:
-            self._conn.close()
+            if self._conn:
-            self._conn = None
+                self._conn.close()
+                self._conn = None
 
     # ================================================================
     # 内部
@@ -339,33 +349,34 @@ class HistoryStore:
         role: str = "",
         extra: Optional[Dict[str, Any]] = None,
     ) -> int:
-        if self._conn is None:
+        with self._conn_lock:
-            return -1
+            if self._conn is None:
-        now = _now_iso()
+                return -1
-        extra_json = json.dumps(extra or {}, ensure_ascii=False)
+            now = _now_iso()
-        cursor = self._conn.execute(
+            extra_json = json.dumps(extra or {}, ensure_ascii=False)
-            """INSERT INTO memory_history
+            cursor = self._conn.execute(
-               (memory_id, isolation_key, event,
+                """INSERT INTO memory_history
-                old_memory, new_memory, old_status, new_status,
+                   (memory_id, isolation_key, event,
-                change_reason, layer, actor_id, role, extra,
+                    old_memory, new_memory, old_status, new_status,
-                created_at, updated_at)
+                    change_reason, layer, actor_id, role, extra,
-               VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+                    created_at, updated_at)
-            (
+                   VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
-                memory_id,
+                (
-                isolation_key,
+                    memory_id,
-                event,
+                    isolation_key,
-                old_memory,
+                    event,
-                new_memory,
+                    old_memory,
-                old_status,
+                    new_memory,
-                new_status,
+                    old_status,
-                change_reason,
+                    new_status,
-                layer,
+                    change_reason,
-                actor_id,
+                    layer,
-                role,
+                    actor_id,
-                extra_json,
+                    role,
-                now,
+                    extra_json,
-                now,
+                    now,
-            ),
+                    now,
-        )
+                ),
-        self._conn.commit()
+            )
-        return cursor.lastrowid
+            self._conn.commit()
+            return cursor.lastrowid

memhub/apps/cli/scripts/prepare-publish-manifest.mjs

@@ -12,7 +12,6 @@
  * It is generated rather than hand-written so the two cannot drift: this reads
  * 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 { dirname, resolve } from "node:path";
@@ -20,12 +19,14 @@ import { fileURLToPath } from "node:url";
 
 const here = dirname(fileURLToPath(import.meta.url));
 const root = resolve(here, "..");
-const repositoryRoot = resolve(root, "../..");
 const require = createRequire(import.meta.url);
-const { resolveBuildVersion } = require("../../../scripts/release-version.cjs");
+const { assertSemver } = require("../../../scripts/release-version.cjs");
-const releaseVersion = resolveBuildVersion(process.env, repositoryRoot);
 const pkgPath = resolve(root, "package.json");
 const pkg = JSON.parse(readFileSync(pkgPath, "utf8"));
+const publishVersion = assertSemver(
+  process.env.MEMHUB_NPM_VERSION?.trim() || pkg.version,
+  "MEMHUB_NPM_VERSION",
+);
 const publishName = process.env.MEMHUB_NPM_PACKAGE?.trim();
 if (!publishName || !/^(?:@[a-z0-9._-]+\/)?[a-z0-9._-]+$/.test(publishName)) {
   throw new Error(
@@ -53,7 +54,7 @@ mkdirSync(outDir, { recursive: true });
 
 const published = { ...pkg };
 published.name = publishName;
-published.version = releaseVersion;
+published.version = publishVersion;
 for (const section of ["dependencies", "devDependencies", "peerDependencies"]) {
   if (!published[section]) continue;
   published[section] = Object.fromEntries(

memhub/apps/cli/src/cli.ts

@@ -331,7 +331,7 @@ import {
   renderVersion,
   renderVersionDocument,
 } from "./usage.js";
-import { CLI_VERSION, PRODUCT } from "./version.js";
+import { CLI_VERSION, PRODUCT, RUNTIME_RELEASE_VERSION } from "./version.js";
 import { runWatchLoop, type WatchEvent, type WatchRead } from "./watch-loop.js";
 
 export interface RunControls {
@@ -417,7 +417,8 @@ async function selectInstallSource(
     throw new Error("release_manifest_url_unavailable");
   }
   const manifestUrl =
-    explicitManifestUrl ?? template.replaceAll("{version}", CLI_VERSION);
+    explicitManifestUrl ??
+    template.replaceAll("{version}", RUNTIME_RELEASE_VERSION);
   const allowedHosts = parseAllowedReleaseHosts(
     environment.MEMHUB_RELEASE_ALLOWED_HOSTS,
   );
@@ -462,7 +463,7 @@ async function resolveRemoteInstallPreview(
 ): Promise<ResolvedRuntimeRelease> {
   return (controls.resolveRuntimeRelease ?? resolveRuntimeRelease)({
     manifestUrl: selection.manifestUrl,
-    expectedVersion: CLI_VERSION,
+    expectedVersion: RUNTIME_RELEASE_VERSION,
     channel: selection.channel,
     platform: process.platform,
     arch: process.arch,
@@ -476,7 +477,7 @@ async function acquireRemoteInstallSource(
 ): Promise<AcquiredRuntimeDistribution> {
   return (controls.acquireRuntimeDistribution ?? acquireRuntimeDistribution)({
     manifestUrl: selection.manifestUrl,
-    expectedVersion: CLI_VERSION,
+    expectedVersion: RUNTIME_RELEASE_VERSION,
     channel: selection.channel,
     platform: process.platform,
     arch: process.arch,
@@ -6520,7 +6521,7 @@ function parseInstallOptions(
         ...(releaseManifestUrl ? { releaseManifestUrl } : {}),
         releaseChannel:
           releaseChannel ??
-          (CLI_VERSION.includes("-") ? "prerelease" : "stable"),
+          (RUNTIME_RELEASE_VERSION.includes("-") ? "prerelease" : "stable"),
         installRoot: resolvedInstallRoot,
         home: resolvedHome,
         uid: uid ?? process.getuid?.(),

memhub/apps/cli/src/publish-manifest.test.ts

@@ -48,7 +48,11 @@ describe("publish manifest is production-safe", () => {
     execFileSync("node", ["scripts/prepare-publish-manifest.mjs"], {
       cwd: root,
       stdio: "pipe",
-      env: { ...process.env, MEMHUB_NPM_PACKAGE: "@example/memhub-cli" },
+      env: {
+        ...process.env,
+        MEMHUB_NPM_PACKAGE: "@example/memhub-cli",
+        MEMHUB_NPM_VERSION: "1.0.5",
+      },
     });
     const published = JSON.parse(
       readFileSync(path.join(root, "dist-publish", "package.json"), "utf8"),
@@ -56,6 +60,7 @@ describe("publish manifest is production-safe", () => {
     expect(published.dependencies).toBeUndefined();
     expect(published.private).toBe(false);
     expect(published.name).toBe("@example/memhub-cli");
+    expect(published.version).toBe("1.0.5");
     expect(published.bin).toMatchObject({
       memhub: "dist/one-memory.cjs",
       "memhub-cli": "dist/one-memory.cjs",

memhub/apps/cli/src/version.ts

@@ -1,5 +1,9 @@
 declare const __ONE_MEMORY_VERSION__: string;
 declare const __ONE_MEMORY_BUILD_NUMBER__: string;
+declare const __MEMHUB_DEFAULT_RUNTIME_VERSION__: string;
+
+const SEMVER =
+  /^(0|[1-9]\d*)\.(0|[1-9]\d*)\.(0|[1-9]\d*)(?:-[0-9A-Za-z-]+(?:\.[0-9A-Za-z-]+)*)?(?:\+[0-9A-Za-z-]+(?:\.[0-9A-Za-z-]+)*)?$/;
 
 const embeddedVersion =
   typeof __ONE_MEMORY_VERSION__ === "string"
@@ -9,13 +13,18 @@ const embeddedBuildNumber =
   typeof __ONE_MEMORY_BUILD_NUMBER__ === "string"
     ? __ONE_MEMORY_BUILD_NUMBER__.trim()
     : "";
+const embeddedRuntimeVersion =
+  typeof __MEMHUB_DEFAULT_RUNTIME_VERSION__ === "string"
+    ? __MEMHUB_DEFAULT_RUNTIME_VERSION__.trim()
+    : "";
 
-export const CLI_VERSION =
+export const CLI_VERSION = SEMVER.test(embeddedVersion)
-  /^(0|[1-9]\d*)\.(0|[1-9]\d*)\.(0|[1-9]\d*)(?:-[0-9A-Za-z-]+(?:\.[0-9A-Za-z-]+)*)?(?:\+[0-9A-Za-z-]+(?:\.[0-9A-Za-z-]+)*)?$/.test(
+  ? embeddedVersion
-    embeddedVersion,
+  : "development";
-  )
+
-    ? embeddedVersion
+export const RUNTIME_RELEASE_VERSION = SEMVER.test(embeddedRuntimeVersion)
-    : "development";
+  ? embeddedRuntimeVersion
+  : CLI_VERSION;
 
 export const CLI_BUILD_NUMBER = /^\d+$/.test(embeddedBuildNumber)
   ? Number(embeddedBuildNumber)

memhub/apps/cli/tsup.config.ts

@@ -2,11 +2,22 @@ import { createRequire } from "node:module";
 import { defineConfig } from "tsup";
 
 const require = createRequire(import.meta.url);
-const { resolveBuildVersion } =
+const { assertSemver, resolveBuildVersion } =
   require("../../scripts/release-version.cjs") as {
+    assertSemver(value: string, label?: string): string;
     resolveBuildVersion(environment?: NodeJS.ProcessEnv, root?: string): string;
   };
 const releaseVersion = resolveBuildVersion(process.env);
+const packageVersion = (require("./package.json") as { version: string })
+  .version;
+const cliVersion = assertSemver(
+  process.env.MEMHUB_NPM_VERSION?.trim() || packageVersion,
+  "MEMHUB_NPM_VERSION",
+);
+const runtimeReleaseVersion = assertSemver(
+  process.env.MEMHUB_DEFAULT_RUNTIME_VERSION?.trim() || releaseVersion,
+  "MEMHUB_DEFAULT_RUNTIME_VERSION",
+);
 
 export default defineConfig({
   entry: { "one-memory": "src/entry.ts" },
@@ -15,7 +26,7 @@ export default defineConfig({
   outDir: "dist",
   clean: true,
   define: {
-    __ONE_MEMORY_VERSION__: JSON.stringify(releaseVersion),
+    __ONE_MEMORY_VERSION__: JSON.stringify(cliVersion),
     __ONE_MEMORY_BUILD_NUMBER__: JSON.stringify(
       process.env.ONE_MEMORY_BUILD_NUMBER ?? "unknown",
     ),
@@ -25,6 +36,7 @@ export default defineConfig({
     __MEMHUB_DEFAULT_RELEASE_MANIFEST_URL_TEMPLATE__: JSON.stringify(
       process.env.MEMHUB_DEFAULT_RELEASE_MANIFEST_URL_TEMPLATE ?? "",
     ),
+    __MEMHUB_DEFAULT_RUNTIME_VERSION__: JSON.stringify(runtimeReleaseVersion),
   },
   // The shared operations package ships TypeScript sources, so it must be
   // inlined into the CJS bundle the same way runtime-control is; otherwise the

memhub/apps/cli/vitest.config.ts

@@ -11,6 +11,7 @@ export default defineConfig({
     __ONE_MEMORY_BUILD_NUMBER__: JSON.stringify("unknown"),
     __ONE_MEMORY_COMMIT__: JSON.stringify("unknown"),
     __MEMHUB_DEFAULT_RELEASE_MANIFEST_URL_TEMPLATE__: JSON.stringify(""),
+    __MEMHUB_DEFAULT_RUNTIME_VERSION__: JSON.stringify(version),
   },
   test: {
     testTimeout: 120_000,

memhub/apps/core/src/one_memory/api.py

@@ -185,7 +185,12 @@ from one_memory.memory_queue import (
     resolve_extraction_state,
 )
 from one_memory.memory_server.config_state import ConfigStore
-from one_memory.memory_tasks import MemoryTask, MemoryTaskStatus, MemoryTaskStore
+from one_memory.memory_tasks import (
+    NON_RETRYABLE_POST_WRITE_CODES,
+    MemoryTask,
+    MemoryTaskStatus,
+    MemoryTaskStore,
+)
 from one_memory.model_presets import seed_preset_profile
 from one_memory.model_probe import ModelProber, ModelProbeResult
 from one_memory.model_profiles import (
@@ -1082,8 +1087,14 @@ def create_app(
             drain_budget = production_drain_budget()
 
             async def reconcile_unknown_writes() -> None:
-                await app.state.memory_processing_supervisor.reconcile_unknown_outputs()
+                await app.state.memory_processing_supervisor.reconcile_unknown_outputs(
-                unknown = await asyncio.to_thread(memory_write_queue.unknown_jobs, limit=128)
+                    exclude_error_codes=NON_RETRYABLE_POST_WRITE_CODES,
+                )
+                unknown = await asyncio.to_thread(
+                    memory_write_queue.unknown_jobs,
+                    limit=128,
+                    exclude_error_codes=NON_RETRYABLE_POST_WRITE_CODES,
+                )
                 for job in unknown:
                     await asyncio.to_thread(
                         memory_write_queue.resolve_unknown_from_linked_output,
@@ -2793,6 +2804,12 @@ def create_app(
         # D75: work queued for the previous model continues with this one.
         # Only tasks that never reached the model are re-bound.
         await asyncio.to_thread(app.state.memory_task_store.rebind_unstarted_profile, profile)
+        # A missing profile is an event-driven wait, not a provider outage.
+        # Release those durable rows only now that activation has changed the
+        # condition; the ordinary drain poll will claim them without a restart.
+        await asyncio.to_thread(
+            app.state.memory_write_queue.readmit_model_profile_waiters
+        )
         # Not awaited: the catch-up reads every conversation ever imported, and
         # the answer to "which profile is active" does not depend on it.
         _start_backfill(app)
@@ -5062,10 +5079,11 @@ def _source_writes_on(app: FastAPI, source_id: str) -> bool:
 
 
 def _sedimentation_claim_enabled(app: FastAPI, source_id: str) -> bool:
-    """Claim queued model work only while both global and source gates allow it."""
+    """Claim queued model work only when policy, source and model all admit it."""
 
     return (
         _config_store(app).sedimentation_enabled()
+        and _model_profile_store(app).active_identity() is not None
         and _history_import_gate_released(app)
         and _source_writes_on(app, source_id)
     )

memhub/apps/core/src/one_memory/imports/live_sync.py

@@ -282,11 +282,24 @@ class PollingLiveSyncSupervisor:
                 and job.status
                 in {ImportStatus.PENDING, ImportStatus.RUNNING, ImportStatus.INTERRUPTED}
             ]
+            bounded_history_incomplete = any(
+                job.source_id == source_id
+                and job.operation == "history"
+                and job.history_after is not None
+                and job.status is not ImportStatus.COMPLETED
+                for job in source_jobs
+            )
             # Establish the realtime EOF independently even while history owns
             # this source. Otherwise the P1 job starts from the P2 checkpoint
             # (or from zero) and becomes a second full-history scan.
             adapter = self.import_supervisor.adapters[source_id]
-            await self._live_cursors(source_id, adapter)
+            live_cursors = await self._live_cursors(source_id, adapter)
+            # A bounded onboarding history is the authority for everything up
+            # to its EOF. An adapter without a native live baseline has no safe
+            # realtime cursor until that history finishes; starting a sync from
+            # zero here would bypass `history_after` and import the old corpus.
+            if bounded_history_incomplete and not live_cursors:
+                return None
             # Only one scan of a source may be in flight: they share the
             # cursor, and two readers means every file read twice. With a
             # catch-up still holding the source there is no free moment to ask

memhub/apps/core/src/one_memory/memory_processing.py

@@ -1522,14 +1522,24 @@ class MemoryProcessingSupervisor:
                 await self._refresh_read_bridge(completed[-1].profile_snapshot)
             return tuple(completed)
 
-    async def reconcile_unknown_outputs(self, *, limit: int = 128) -> int:
+    async def reconcile_unknown_outputs(
+        self,
+        *,
+        limit: int = 128,
+        exclude_error_codes: tuple[str, ...] = (),
+    ) -> int:
         """Publish HY outputs that can be proven by an unknown task's request id.
 
         This inquiry is read-only on HY and never invokes the model. Rows with
         no exact match remain unknown for the person's explicit retry choice.
         """
         reconciled = 0
-        for task in await in_the_background(self.tasks.unknown_tasks, limit=limit):
+        tasks = await in_the_background(
+            self.tasks.unknown_tasks,
+            limit=limit,
+            exclude_error_codes=exclude_error_codes,
+        )
+        for task in tasks:
             runner = self.runner_factory()
             bridge: MemoryBridgeProtocol | None = None
             try:

memhub/apps/core/src/one_memory/memory_server/durable_queue.py

@@ -38,10 +38,11 @@ from one_memory.memory_server.migrations import memory_write_jobs_sqlite_schema
 from one_memory.memory_server.queue import build_reclaim_sqlite_statements
 from one_memory.memory_server.registry import WriteRecord
 from one_memory.memory_tasks import (
-    POLICY_STOPPED_CODES as _TASK_POLICY_STOPPED_CODES,
+    NON_RETRYABLE_POST_WRITE_CODES,
+    MemoryTaskStatus,
 )
 from one_memory.memory_tasks import (
-    MemoryTaskStatus,
+    POLICY_STOPPED_CODES as _TASK_POLICY_STOPPED_CODES,
 )
 from one_memory.service_levels import ServiceLevel, service_level_context
 
@@ -1404,11 +1405,12 @@ class DurableWriteQueue:
     ) -> int:
         """Re-admit `waiting_for_model` jobs whose wait has elapsed.
 
-        A model outage clears on its own, so this state is retried on a timer
+        A model outage clears on its own, so transient waiters retry on a timer
         rather than waiting for a grant event like `waiting_for_permission`.
-        Keyed on `updated_at` so repeated calls while a job is still parked do
+        `model_profile_not_active` is excluded: only explicit activation can
-        not reset its own clock; the attempt consumed by the parking claim is
+        change that condition. Keyed on `updated_at` so repeated calls while a
-        returned, the same way `readmit_parked` does.
+        job is still parked do not reset its own clock; the attempt consumed by
+        the parking claim is returned, the same way `readmit_parked` does.
         """
         now = now or datetime.now(UTC)
         threshold = (now - after).isoformat()
@@ -1417,8 +1419,33 @@ class DurableWriteQueue:
                 "UPDATE memory_write_jobs SET state=?, lease_owner=NULL, "
                 "lease_expires_at=NULL, attempts=CASE WHEN attempts>0 "
                 "THEN attempts-1 ELSE 0 END, last_error=NULL, updated_at=? "
-                "WHERE state=? AND updated_at <= ? AND hy_invoked=0",
+                "WHERE state=? AND updated_at <= ? AND hy_invoked=0 "
-                (PENDING, now.isoformat(), WAITING_FOR_MODEL, threshold),
+                "AND (last_error IS NULL OR last_error<>?)",
+                (
+                    PENDING,
+                    now.isoformat(),
+                    WAITING_FOR_MODEL,
+                    threshold,
+                    "model_profile_not_active",
+                ),
+            )
+        return changed.rowcount
+
+    def readmit_model_profile_waiters(self) -> int:
+        """Re-admit no-profile work only after model activation changes the condition."""
+        now = datetime.now(UTC).isoformat()
+        with main_database(self.database_path).write() as connection, connection:
+            changed = connection.execute(
+                "UPDATE memory_write_jobs SET state=?, lease_owner=NULL, "
+                "lease_expires_at=NULL, attempts=CASE WHEN attempts>0 "
+                "THEN attempts-1 ELSE 0 END, last_error=NULL, updated_at=? "
+                "WHERE state=? AND last_error=? AND hy_invoked=0",
+                (
+                    PENDING,
+                    now,
+                    WAITING_FOR_MODEL,
+                    "model_profile_not_active",
+                ),
             )
         return changed.rowcount
 
@@ -1516,9 +1543,13 @@ class DurableWriteQueue:
             connection,
         ):
             row = connection.execute(
-                "SELECT state FROM memory_write_jobs WHERE job_id=?", (job_id,)
+                "SELECT state,last_error FROM memory_write_jobs WHERE job_id=?", (job_id,)
             ).fetchone()
-            if row is None or str(row["state"]) != UNKNOWN:
+            if (
+                row is None
+                or str(row["state"]) != UNKNOWN
+                or str(row["last_error"]) in NON_RETRYABLE_POST_WRITE_CODES
+            ):
                 return 0
             connection.execute(
                 """
@@ -1675,8 +1706,15 @@ class DurableWriteQueue:
             rows = connection.execute(
                 "SELECT job_id FROM memory_write_jobs "
                 "WHERE state=? AND attempts=1 AND attempts < max_attempts "
+                f"AND COALESCE(last_error, '') NOT IN "
+                f"({','.join('?' for _ in NON_RETRYABLE_POST_WRITE_CODES)}) "
                 "AND updated_at <= ? ORDER BY updated_at, job_id LIMIT ?",
-                (UNKNOWN, older_than.isoformat(), limit),
+                (
+                    UNKNOWN,
+                    *NON_RETRYABLE_POST_WRITE_CODES,
+                    older_than.isoformat(),
+                    limit,
+                ),
             ).fetchall()
             job_ids = tuple(str(row["job_id"]) for row in rows)
             if not job_ids:
@@ -1740,6 +1778,7 @@ class DurableWriteQueue:
         *,
         limit: int = 128,
         after_cursor: tuple[str, str] | None = None,
+        exclude_error_codes: tuple[str, ...] = (),
     ) -> tuple[UnknownWriteJob, ...]:
         """List unknown rows that still need an inquiry, in stable order.
 
@@ -1756,6 +1795,9 @@ class DurableWriteQueue:
         """
         if type(limit) is not int or limit < 1:
             raise ValueError("limit must be positive")
+        excluded = tuple(dict.fromkeys(exclude_error_codes))
+        if not all(isinstance(code, str) and code for code in excluded):
+            raise ValueError("exclude_error_codes must contain non-empty strings")
         cursor_clause = ""
         params: tuple[Any, ...] = ()
         if after_cursor is not None:
@@ -1768,14 +1810,20 @@ class DurableWriteQueue:
             updated_at, job_id = after_cursor
             cursor_clause = "AND (updated_at, job_id) > (?, ?)"
             params = (updated_at, job_id)
+        if excluded:
+            placeholders = ",".join("?" for _ in excluded)
+            excluded_clause = f"AND COALESCE(last_error, '') NOT IN ({placeholders})"
+        else:
+            excluded_clause = ""
         with main_database(self.database_path).read(row_factory=sqlite3.Row) as connection:
             rows = connection.execute(
                 "SELECT job_id, idempotency_key, created_at, updated_at, last_error "
                 "FROM memory_write_jobs "
                 "WHERE state=? "
                 f"{cursor_clause} "
+                f"{excluded_clause} "
                 "ORDER BY updated_at, job_id LIMIT ?",
-                (UNKNOWN, *params, limit),
+                (UNKNOWN, *params, *excluded, limit),
             ).fetchall()
         return tuple(
             UnknownWriteJob(
@@ -2071,7 +2119,12 @@ async def execute_job(
         # Never `logger.exception` here: it prints the exception's message,
         # which may be a provider response or the user's own text. Log the
         # already-sanitised code instead, matching `hy-failure-classification`.
-        logger.error("durable_write_job_failed job_id=%s error_code=%s", job_id, code)
+        if code == "model_profile_not_active":
+            logger.debug("durable_write_job_parked job_id=%s error_code=%s", job_id, code)
+        elif state in PARKED_STATES:
+            logger.info("durable_write_job_parked job_id=%s error_code=%s", job_id, code)
+        else:
+            logger.error("durable_write_job_failed job_id=%s error_code=%s", job_id, code)
         return
     await in_the_background(queue.settle, job_id, state=COMPLETED, owner=owner)
 

memhub/apps/core/src/one_memory/memory_tasks.py

@@ -67,6 +67,12 @@ POLICY_STOPPED_CODES = (
     "hy_memory_evidence_changed",
 )
 
+#: HY completed a write, but Core could not prove that the selected mutation
+#: target belonged to this task's frozen authority. Reconciliation may inspect
+#: HY by request id, but neither automatic nor user-confirmed generic retry may
+#: invoke the model again for these codes.
+NON_RETRYABLE_POST_WRITE_CODES = ("hy_memory_write_verification_authority_invalid",)
+
 
 class MemoryTask(BaseModel):
     model_config = ConfigDict(frozen=True, extra="forbid")
@@ -116,6 +122,24 @@ PROFILE_STOP_CODES = (
 ACTIVE_WINDOW_DEFER_LIMIT = 2048
 
 
+def _task_only_unknown_clause(connection: sqlite3.Connection) -> str:
+    """Exclude tasks whose Evidence is already owned by the durable queue."""
+    linked_table = connection.execute(
+        "SELECT 1 FROM sqlite_master WHERE type='table' AND name='memory_write_job_evidence'"
+    ).fetchone()
+    if linked_table is None:
+        return ""
+    return """
+        AND NOT EXISTS (
+            SELECT 1
+            FROM memory_task_inputs AS inputs, json_each(inputs.refs_json) AS ref
+            JOIN memory_write_job_evidence AS linked
+              ON linked.evidence_id=json_extract(ref.value, '$.evidence_id')
+            WHERE inputs.task_id=memory_tasks.task_id
+        )
+    """
+
+
 @dataclass(frozen=True, slots=True)
 class ActiveMemoryTask:
     """One task plus only the evidence fields the scheduler needs."""
@@ -1314,7 +1338,7 @@ class MemoryTaskStore:
         live_clause = f"task.task_id IN ({live_placeholders})" if live_ids else "0"
         not_live_clause = f"task.task_id NOT IN ({live_placeholders})" if live_ids else "1"
         excluded_clause = f"task.task_id NOT IN ({excluded_placeholders})" if excluded else "1"
-        common = f"""
+        common_base = f"""
             task.status IN (?,?,?,?)
             AND coalesce(task.last_error_code, '') NOT IN (
                 'hy_memory_input_exceeds_window', 'model_content_outcome_unknown'
@@ -1333,7 +1357,33 @@ class MemoryTaskStore:
         """
         moment = "coalesce(evidence.created_at, evidence.imported_at)"
 
+        def common_for(connection: sqlite3.Connection) -> str:
+            has_durable_queue = connection.execute(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type='table' "
+                "AND name IN ('memory_write_jobs','memory_write_job_evidence')"
+            ).fetchone()[0]
+            if int(has_durable_queue) != 2:
+                return common_base
+            return (
+                common_base
+                + """
+                AND NOT EXISTS (
+                    SELECT 1
+                    FROM memory_task_inputs AS inputs, json_each(inputs.refs_json) AS ref
+                    JOIN memory_write_job_evidence AS linked
+                      ON linked.evidence_id=json_extract(ref.value, '$.evidence_id')
+                    JOIN memory_write_jobs AS job ON job.job_id=linked.job_id
+                    WHERE inputs.task_id=task.task_id
+                      AND job.state IN (
+                          'pending', 'running', 'unknown',
+                          'waiting_for_permission', 'waiting_for_model'
+                      )
+                )
+            """
+            )
+
         def run(connection: sqlite3.Connection, *, defer: bool) -> list[sqlite3.Row]:
+            common = common_for(connection)
             in_window = (
                 "1"
                 if extraction_since is None
@@ -1430,6 +1480,7 @@ class MemoryTaskStore:
         def has_defer_candidate(connection: sqlite3.Connection) -> bool:
             if extraction_since is None:
                 return False
+            common = common_for(connection)
             row = connection.execute(
                 f"""
                 SELECT 1
@@ -2146,6 +2197,7 @@ class MemoryTaskStore:
         *,
         limit: int = 128,
         after_cursor: tuple[str, str] | None = None,
+        exclude_error_codes: tuple[str, ...] = (),
     ) -> tuple[MemoryTask, ...]:
         """List `unknown` tasks that still need an external inquiry.
 
@@ -2160,6 +2212,9 @@ class MemoryTaskStore:
         """
         if type(limit) is not int or limit < 1:
             raise ValueError("limit must be positive")
+        excluded = tuple(dict.fromkeys(exclude_error_codes))
+        if not all(isinstance(code, str) and code for code in excluded):
+            raise ValueError("exclude_error_codes must contain non-empty strings")
         if after_cursor is not None:
             if (
                 not isinstance(after_cursor, tuple)
@@ -2173,13 +2228,19 @@ class MemoryTaskStore:
         else:
             cursor_clause = ""
             params = ()
+        if excluded:
+            placeholders = ",".join("?" for _ in excluded)
+            excluded_clause = f"AND COALESCE(last_error_code, '') NOT IN ({placeholders})"
+        else:
+            excluded_clause = ""
         with self._connect() as connection:
             rows = connection.execute(
                 "SELECT * FROM memory_tasks "
                 "WHERE status=? "
                 f"{cursor_clause} "
+                f"{excluded_clause} "
                 "ORDER BY updated_at, task_id LIMIT ?",
-                (MemoryTaskStatus.UNKNOWN, *params, limit),
+                (MemoryTaskStatus.UNKNOWN, *params, *excluded, limit),
             ).fetchall()
         return tuple(self._to_task(row) for row in rows)
 
@@ -2211,20 +2272,26 @@ class MemoryTaskStore:
         """
         now = datetime.now(UTC).isoformat()
         with self._connect(writing=True) as connection:
+            task_only = _task_only_unknown_clause(connection)
+            blocked = ",".join("?" for _ in NON_RETRYABLE_POST_WRITE_CODES)
             changed = connection.execute(
-                """
+                f"""
                 UPDATE memory_tasks
                 SET status=?, runner_token=NULL, attempts=0,
                     last_error_code='model_content_manual_retry_requested', updated_at=?
                 WHERE task_id IN (
                     SELECT task_id FROM memory_tasks
-                    WHERE status=? ORDER BY updated_at, task_id LIMIT ?
+                    WHERE status=?
+                      AND COALESCE(last_error_code, '') NOT IN ({blocked})
+                      {task_only}
+                    ORDER BY updated_at, task_id LIMIT ?
                 )
                 """,
                 (
                     MemoryTaskStatus.PENDING,
                     now,
                     MemoryTaskStatus.UNKNOWN,
+                    *NON_RETRYABLE_POST_WRITE_CODES,
                     limit,
                 ),
             )
@@ -2246,14 +2313,18 @@ class MemoryTaskStore:
             raise ValueError("limit must be positive")
         now = datetime.now(UTC).isoformat()
         with self._connect(writing=True) as connection:
+            task_only = _task_only_unknown_clause(connection)
+            blocked = ",".join("?" for _ in NON_RETRYABLE_POST_WRITE_CODES)
             changed = connection.execute(
-                """
+                f"""
                 UPDATE memory_tasks
                 SET status=?, runner_token=NULL,
                     last_error_code='model_content_unknown_auto_retry', updated_at=?
                 WHERE task_id IN (
                     SELECT task_id FROM memory_tasks
                     WHERE status=? AND attempts=1 AND updated_at <= ?
+                      AND COALESCE(last_error_code, '') NOT IN ({blocked})
+                      {task_only}
                     ORDER BY updated_at, task_id LIMIT ?
                 )
                 """,
@@ -2262,6 +2333,7 @@ class MemoryTaskStore:
                     now,
                     MemoryTaskStatus.UNKNOWN,
                     older_than.isoformat(),
+                    *NON_RETRYABLE_POST_WRITE_CODES,
                     limit,
                 ),
             )

memhub/apps/core/src/one_memory/memory_visibility.py

@@ -200,10 +200,12 @@ def _trigger_sql() -> list[tuple[str, str]]:
         CREATE TRIGGER IF NOT EXISTS memory_visibility_dirty_on_link_update
         AFTER UPDATE ON governed_memory_evidence
         BEGIN
-            INSERT OR IGNORE INTO memory_visibility_dirty (memory_id)
+            INSERT INTO memory_visibility_dirty (memory_id)
-            VALUES (NEW.memory_id);
+            VALUES (NEW.memory_id)
-            INSERT OR IGNORE INTO memory_visibility_dirty (memory_id)
+            ON CONFLICT(memory_id) DO NOTHING;
-            VALUES (OLD.memory_id);
+            INSERT INTO memory_visibility_dirty (memory_id)
+            VALUES (OLD.memory_id)
+            ON CONFLICT(memory_id) DO NOTHING;
         END;""",
         ),
         (
@@ -294,6 +296,10 @@ def install_triggers(db: sqlite3.Connection) -> None:
         str(row["name"])
         for row in cursor.execute("SELECT name FROM sqlite_master WHERE type='table'")
     }
+    # CREATE TRIGGER IF NOT EXISTS cannot repair the legacy definition. Replace
+    # this derived-cache trigger transactionally on every startup so old
+    # libraries get the row-level conflict handling without a schema migration.
+    cursor.execute("DROP TRIGGER IF EXISTS memory_visibility_dirty_on_link_update")
     for table, statement in _trigger_sql():
         if table not in present:
             continue
@@ -303,11 +309,15 @@ def install_triggers(db: sqlite3.Connection) -> None:
 def startup(database_path: Path) -> None:
     """Idempotent: create the derived tables and triggers. No version change.
 
-    Runs through the write connection because the triggers are schema. It is
+    Runs through the write connection because the triggers are schema. The
-    one short transaction and happens once per start.
+    trigger replacement is one short transaction and happens once per start.
     """
-    with main_database(database_path).write() as db:
+    with main_database(database_path).write() as db, db:
         ensure_tables(db)
+        # ensure_tables uses executescript(), which commits any pending legacy
+        # transaction. Begin only afterward so replacing the installed trigger
+        # is one atomic schema change and a failed CREATE restores the old one.
+        db.execute("BEGIN IMMEDIATE")
         install_triggers(db)
 
 

memhub/apps/core/tests/test_live_sync.py

@@ -94,6 +94,67 @@ class SettledFixtureAdapter(IncrementalFixtureAdapter):
         }
 
 
+class BoundedOnboardingAdapter:
+    """No native baseline: bounded history must establish the live cursor."""
+
+    descriptor = IncrementalFixtureAdapter.descriptor
+
+    def __init__(self) -> None:
+        self.include_new = False
+
+    async def detect(self) -> bool:
+        return True
+
+    async def scan(self, options: ScanOptions):
+        current = options.cursors.get("e" * 64)
+        checkpoint = current.checkpoint if current is not None else None
+        if checkpoint not in {"byte:10", "byte:20"}:
+            yield ScannedEvidence(
+                evidence=EvidenceInput(
+                    source_id="fixture",
+                    conversation_id="old-conversation",
+                    message_id="old-message",
+                    role="user",
+                    written_by=WrittenBy.PERSON,
+                    created_at=datetime(2020, 1, 1, tzinfo=UTC),
+                    verbatim="old history",
+                    origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+                ),
+                cursor=SourceCursor(
+                    file_key="e" * 64,
+                    file_identity="bounded",
+                    checkpoint="byte:10",
+                    checkpoint_anchor="f" * 64,
+                    observed_size=20,
+                    observed_modified_ns=1,
+                ),
+            )
+            checkpoint = "byte:10"
+        if self.include_new and checkpoint != "byte:20":
+            yield ScannedEvidence(
+                evidence=EvidenceInput(
+                    source_id="fixture",
+                    conversation_id="new-conversation",
+                    message_id="new-message",
+                    role="user",
+                    written_by=WrittenBy.PERSON,
+                    created_at=datetime(2026, 10, 18, tzinfo=UTC),
+                    verbatim="new append",
+                    origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+                ),
+                cursor=SourceCursor(
+                    file_key="e" * 64,
+                    file_identity="bounded",
+                    checkpoint="byte:20",
+                    checkpoint_anchor="1" * 64,
+                    observed_size=20,
+                    observed_modified_ns=2,
+                    read_to_end=True,
+                ),
+            )
+        yield ScanProgressEvent(phase=ScanPhase.DONE)
+
+
 class HoldingHistoryAdapter:
     """First scan blocks until cancelled; later scans yield one new record."""
 
@@ -328,6 +389,56 @@ def test_live_sync_does_not_resume_a_history_import_the_user_paused(tmp_path) ->
     asyncio.run(scenario())
 
 
+def test_bounded_onboarding_history_establishes_cursor_before_first_sync(tmp_path) -> None:
+    async def scenario() -> None:
+        database = tmp_path / "one-memory.sqlite3"
+        evidence = EvidenceStore(database)
+        jobs = ImportJobStore(database)
+        adapter = BoundedOnboardingAdapter()
+        imports = ImportSupervisor(
+            adapters={"fixture": adapter},
+            evidence_store=evidence,
+            job_store=jobs,
+        )
+        live_sync = PollingLiveSyncSupervisor(
+            source_ids=("fixture",),
+            import_supervisor=imports,
+            job_store=jobs,
+        )
+        history = imports.create_job("fixture", operation="history")
+        with sqlite3.connect(database) as connection, connection:
+            connection.execute(
+                "UPDATE import_jobs SET history_after=? WHERE job_id=?",
+                (datetime(2026, 10, 15, tzinfo=UTC).isoformat(), history.job_id),
+            )
+
+        # With no native baseline, sync must wait instead of scanning from zero
+        # and bypassing the onboarding history window.
+        assert await live_sync.trigger("fixture") is None
+        assert not [job for job in jobs.list_jobs() if job.operation == "sync"]
+
+        imports.enqueue(history.job_id)
+        completed = await imports.wait(history.job_id)
+        assert completed.status is ImportStatus.COMPLETED
+        assert completed.stored == 0
+        assert completed.skipped == 1
+        assert evidence.count() == 0
+
+        # Once bounded history has advanced the cursor to EOF, normal realtime
+        # appends are admitted without retaining the three-day cutoff forever.
+        adapter.include_new = True
+        synced = await live_sync.trigger("fixture")
+        assert synced is not None
+        assert synced.status is ImportStatus.COMPLETED
+        assert synced.stored == 1
+        records = evidence.list_extractable()
+        assert [record.verbatim for record in records] == ["new append"]
+        await live_sync.shutdown()
+        await imports.shutdown()
+
+    asyncio.run(scenario())
+
+
 def test_a_sync_left_running_after_its_worker_died_is_started_again(tmp_path) -> None:
     """A running row with no live worker used to be returned as 'already syncing'.
 

memhub/apps/core/tests/test_memory_processing.py

@@ -4,7 +4,7 @@ import logging
 import sqlite3
 import threading
 import time
-from datetime import UTC, datetime
+from datetime import UTC, datetime, timedelta
 
 import pytest
 
@@ -56,7 +56,12 @@ from one_memory.memory_processing import (
 )
 from one_memory.memory_queue import classify
 from one_memory.memory_server.config_state import ConfigStore
-from one_memory.memory_tasks import MemoryTaskKind, MemoryTaskStatus, MemoryTaskStore
+from one_memory.memory_tasks import (
+    NON_RETRYABLE_POST_WRITE_CODES,
+    MemoryTaskKind,
+    MemoryTaskStatus,
+    MemoryTaskStore,
+)
 from one_memory.model_profiles import (
     ModelCapabilities,
     ModelProfileInput,
@@ -2613,6 +2618,123 @@ def test_an_unknown_task_without_inputs_does_not_end_the_reconcile_pass(tmp_path
     assert [memory.memory_id for memory in governance.list(user_id="local")] == ["hy-recovered"]
 
 
+def test_sticky_authority_unknowns_do_not_starve_ordinary_reconciliation(tmp_path) -> None:
+    stores = _setup(tmp_path)
+    evidence, profiles, tasks, governance, secrets, profile = stores
+    record = _put(
+        evidence,
+        message_id="fair-unknown",
+        role="user",
+        written_by=WrittenBy.PERSON,
+        content="ordinary unknown remains reconcilable",
+    )
+    sticky = [
+        tasks.create(
+            request_id=f"sticky-authority-{index}",
+            user_id=record.user_id,
+            kind=MemoryTaskKind.EXTRACT,
+            input_ref=record.evidence_id,
+            profile=profile,
+            input_records=(record,),
+        )
+        for index in range(129)
+    ]
+    ordinary = tasks.create(
+        request_id="ordinary-unknown-after-sticky",
+        user_id=record.user_id,
+        kind=MemoryTaskKind.EXTRACT,
+        input_ref=record.evidence_id,
+        profile=profile,
+        input_records=(record,),
+    )
+    anchor = datetime(2026, 1, 1, tzinfo=UTC)
+    with sqlite3.connect(tasks.database_path) as connection, connection:
+        connection.executescript(
+            """
+            CREATE TABLE source_grants (
+                user_id TEXT NOT NULL, source_id TEXT NOT NULL,
+                root_path TEXT NOT NULL, candidate_kind TEXT NOT NULL,
+                granted_at TEXT NOT NULL, revoked_at TEXT,
+                PRIMARY KEY (user_id, source_id)
+            );
+            """
+        )
+        connection.execute(
+            "INSERT INTO source_grants VALUES (?,?,?,?,?,NULL)",
+            (
+                record.user_id,
+                record.source_id,
+                "/authorized",
+                "default",
+                datetime.now(UTC).isoformat(),
+            ),
+        )
+        for index, task in enumerate(sticky):
+            connection.execute(
+                "UPDATE memory_tasks SET status='unknown',last_error_code=?,updated_at=? "
+                "WHERE task_id=?",
+                (
+                    "hy_memory_write_verification_authority_invalid",
+                    (anchor + timedelta(seconds=index)).isoformat(),
+                    task.task_id,
+                ),
+            )
+        connection.execute(
+            "UPDATE memory_tasks SET status='unknown',last_error_code=?,updated_at=? "
+            "WHERE task_id=?",
+            (
+                "model_content_outcome_unknown",
+                (anchor + timedelta(minutes=3)).isoformat(),
+                ordinary.task_id,
+            ),
+        )
+
+    class InquiryOnlyBridge(FakeBridge):
+        def __init__(self) -> None:
+            super().__init__(
+                outputs=(
+                    HyMemoryOutput(
+                        memory_id="ordinary-recovered",
+                        content="ordinary unknown remains reconcilable",
+                        layer="L2_FACT",
+                        owner="user",
+                        evidence_ids=(record.evidence_id,),
+                    ),
+                )
+            )
+            self.inquiries: list[str] = []
+
+        async def outputs_for_request(self, **kwargs) -> tuple[HyMemoryOutput, ...]:
+            self.inquiries.append(str(kwargs["request_id"]))
+            return self.outputs or ()
+
+    bridge = InquiryOnlyBridge()
+    supervisor = MemoryProcessingSupervisor(
+        tasks=tasks,
+        profiles=profiles,
+        secrets=secrets,
+        evidence=evidence,
+        governance=governance,
+        data_dir=tmp_path,
+        runner_factory=lambda: _runner(tmp_path, stores, FakeBridge()),
+        bridge_factory=lambda _profile, _secret, _data_dir: bridge,
+    )
+
+    assert (
+        asyncio.run(
+            supervisor.reconcile_unknown_outputs(
+                limit=128,
+                exclude_error_codes=NON_RETRYABLE_POST_WRITE_CODES,
+            )
+        )
+        == 1
+    )
+    assert bridge.inquiries == ["ordinary-unknown-after-sticky"]
+    assert bridge.requests == []
+    assert tasks.require(ordinary.task_id).status is MemoryTaskStatus.COMPLETED
+    assert all(tasks.require(task.task_id).status is MemoryTaskStatus.UNKNOWN for task in sticky)
+
+
 def test_two_workspaces_with_one_conversation_id_prepare_apart(tmp_path) -> None:
     """Cursor can reuse a conversation id across projects.
 
@@ -3441,6 +3563,58 @@ def test_digest_scopes_separate_applications_and_dates(tmp_path) -> None:
     assert len({_hy_agent_id((record,)) for record in records}) == 3
 
 
+def test_workspace_requests_carry_conversation_authority_beyond_agent_scope(tmp_path) -> None:
+    """Same-day workspace scopes collide by design; session_id must separate writes."""
+    evidence, _profiles, tasks, _governance, _secrets, profile = _setup(tmp_path)
+    first = _put(
+        evidence,
+        message_id="conversation-a",
+        role="user",
+        written_by=WrittenBy.PERSON,
+        content="same workspace, first conversation",
+        conversation_id="conversation-a",
+        workspace="repo-a",
+        created_at=datetime(2026, 10, 18, 8, tzinfo=UTC),
+    )
+    second = _put(
+        evidence,
+        message_id="conversation-b",
+        role="user",
+        written_by=WrittenBy.PERSON,
+        content="same workspace, second conversation",
+        conversation_id="conversation-b",
+        workspace="repo-a",
+        created_at=datetime(2026, 10, 18, 9, tzinfo=UTC),
+    )
+    next_day = _put(
+        evidence,
+        message_id="conversation-a-next-day",
+        role="user",
+        written_by=WrittenBy.PERSON,
+        content="same conversation on the next day",
+        conversation_id="conversation-a",
+        workspace="repo-a",
+        created_at=datetime(2026, 10, 19, 8, tzinfo=UTC),
+    )
+
+    requests = []
+    for record in (first, second, next_day):
+        task = tasks.create(
+            request_id=f"authority:{record.evidence_id}",
+            user_id=record.user_id,
+            kind=MemoryTaskKind.EXTRACT,
+            input_ref=record.evidence_id,
+            profile=profile,
+            input_records=(record,),
+        )
+        requests.append(MemoryTaskRunner._request(task, (record,)))
+
+    assert requests[0].agent_id == requests[1].agent_id
+    assert requests[0].session_id == "conversation-a"
+    assert requests[1].session_id == "conversation-b"
+    assert requests[2].agent_id != requests[0].agent_id
+
+
 def test_cancelled_multi_source_lock_acquisition_releases_earlier_source(tmp_path) -> None:
     evidence, profiles, tasks, governance, secrets, _profile = _setup(tmp_path)
     first = _put(

memhub/apps/core/tests/test_memory_server_durable_queue.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 import asyncio
 import contextlib
 import json
+import logging
 import sqlite3
 import threading
 import time
@@ -1369,24 +1370,30 @@ def test_a_machine_error_code_is_preserved(tmp_path) -> None:
     assert safe_error_code(ValueError("some sentence with spaces")) == "ValueError"
 
 
-def test_an_inactive_profile_parks_for_model_before_any_send(tmp_path) -> None:
+def test_an_inactive_profile_parks_without_error_logging(tmp_path, caplog) -> None:
     """A pre-bridge model-not-ready code parks and keeps the intent."""
     path = tmp_path / "one-memory.sqlite3"
     queue = DurableWriteQueue(path)
     record = queue.submit(_request())
     _claim_through_drain(queue, record.job_id)
 
-    asyncio.run(
+    with caplog.at_level(
-        execute_job(
+        logging.DEBUG,
-            record.job_id,
+        logger="one_memory.memory_server.durable_queue",
-            queue=queue,
+    ):
-            supervisor=_Recorder(error=RuntimeError("model_profile_not_active")),
+        asyncio.run(
-            evidence=_EvidenceStore({"e1"}),
+            execute_job(
+                record.job_id,
+                queue=queue,
+                supervisor=_Recorder(error=RuntimeError("model_profile_not_active")),
+                evidence=_EvidenceStore({"e1"}),
+            )
         )
-    )
 
     assert _state(path, record.job_id) == WAITING_FOR_MODEL
     assert _rows(path)[0]["hy_invoked"] == 0
+    assert "durable_write_job_failed" not in caplog.text
+    assert "durable_write_job_parked" in caplog.text
 
 
 def test_an_authorization_stop_cancels_instead_of_failing(tmp_path) -> None:
@@ -1546,7 +1553,7 @@ def test_parking_does_not_burn_the_retry_budget(tmp_path) -> None:
     assert queue.get("idem-1").state == PENDING
 
 
-def test_a_model_waiter_is_readmitted_once_its_wait_elapses(tmp_path) -> None:
+def test_a_transient_model_waiter_is_readmitted_once_its_wait_elapses(tmp_path) -> None:
     """A model outage clears on its own; nothing re-admits it otherwise.
 
     `waiting_for_permission` needs a grant event, but `waiting_for_model` is
@@ -1564,7 +1571,7 @@ def test_a_model_waiter_is_readmitted_once_its_wait_elapses(tmp_path) -> None:
     queue.settle(
         record.job_id,
         state=WAITING_FOR_MODEL,
-        error_code="model_profile_not_active",
+        error_code="model_secret_unavailable",
         hy_invoked=False,
     )
 
@@ -1581,8 +1588,8 @@ def test_a_model_waiter_is_readmitted_once_its_wait_elapses(tmp_path) -> None:
     assert _rows(path)[0]["attempts"] == 0
 
 
-def test_readmitting_a_model_waiter_clears_its_stale_detail(tmp_path) -> None:
+def test_no_profile_waiter_is_readmitted_only_after_activation(tmp_path) -> None:
-    """The timer readmission path must clear the old blocking reason too."""
+    """No timer can create a model profile; activation owns this transition."""
     path = tmp_path / "one-memory.sqlite3"
     queue = DurableWriteQueue(path)
     record = queue.submit(_request())
@@ -1594,13 +1601,50 @@ def test_readmitting_a_model_waiter_clears_its_stale_detail(tmp_path) -> None:
     )
     now = datetime.now(UTC)
 
-    assert queue.readmit_model_waiters(now=now + timedelta(minutes=2)) == 1
+    assert queue.readmit_model_waiters(now=now + timedelta(minutes=2)) == 0
+    parked = queue.get("idem-1")
+    assert parked.state == WAITING_FOR_MODEL
+    assert parked.detail == "model_profile_not_active"
 
+    assert queue.readmit_model_profile_waiters() == 1
     readmitted = queue.get("idem-1")
     assert readmitted.state == PENDING
     assert readmitted.detail == ""
 
 
+def test_no_profile_backlog_stays_parked_across_maintenance_cycles(tmp_path) -> None:
+    path = tmp_path / "one-memory.sqlite3"
+    queue = DurableWriteQueue(path)
+    for index in range(100):
+        record = queue.submit(_request(key=f"no-profile-{index}"))
+        queue.settle(
+            record.job_id,
+            state=WAITING_FOR_MODEL,
+            error_code="model_profile_not_active",
+            hy_invoked=False,
+        )
+    with sqlite3.connect(path) as connection, connection:
+        connection.execute(
+            "UPDATE memory_write_jobs SET attempts=1,"
+            "updated_at='2000-01-01T00:00:00+00:00'"
+        )
+
+    first = datetime.now(UTC)
+    assert queue.readmit_model_waiters(now=first, after=timedelta(0)) == 0
+    assert (
+        queue.readmit_model_waiters(
+            now=first + timedelta(minutes=2),
+            after=timedelta(0),
+        )
+        == 0
+    )
+    with sqlite3.connect(path) as connection:
+        assert connection.execute(
+            "SELECT state,attempts,COUNT(*) FROM memory_write_jobs "
+            "GROUP BY state,attempts"
+        ).fetchall() == [(WAITING_FOR_MODEL, 1, 100)]
+
+
 def test_readmit_parked_cannot_resurrect_a_terminal_row(tmp_path) -> None:
     """Only parked states may be re-admitted, never a finished write.
 
@@ -3472,6 +3516,41 @@ def test_unknown_jobs_does_not_resolve_or_make_a_row_claimable(tmp_path) -> None
     assert queue.resolve_unknown(found[0].job_id, found=False, detail="not in HY") == DEAD
 
 
+def test_unknown_jobs_can_exclude_sticky_authority_failures_before_limit(tmp_path) -> None:
+    path = tmp_path / "one-memory.sqlite3"
+    queue = DurableWriteQueue(path)
+    sticky = [
+        queue.submit(_request(key=f"sticky-{index}", evidence=(f"sticky-e{index}",)))
+        for index in range(129)
+    ]
+    ordinary = queue.submit(_request(key="ordinary", evidence=("ordinary-e",)))
+    anchor = datetime(2026, 1, 1, tzinfo=UTC)
+    with sqlite3.connect(path) as connection, connection:
+        for index, record in enumerate(sticky):
+            connection.execute(
+                "UPDATE memory_write_jobs SET state='unknown',attempts=1,hy_invoked=1,"
+                "last_error=?,updated_at=? WHERE job_id=?",
+                (
+                    "hy_memory_write_verification_authority_invalid",
+                    (anchor + timedelta(seconds=index)).isoformat(),
+                    record.job_id,
+                ),
+            )
+        connection.execute(
+            "UPDATE memory_write_jobs SET state='unknown',attempts=1,hy_invoked=1,"
+            "last_error='model_content_outcome_unknown',updated_at=? WHERE job_id=?",
+            ((anchor + timedelta(minutes=3)).isoformat(), ordinary.job_id),
+        )
+
+    found = queue.unknown_jobs(
+        limit=128,
+        exclude_error_codes=("hy_memory_write_verification_authority_invalid",),
+    )
+
+    assert [job.job_id for job in found] == [ordinary.job_id]
+    assert all(queue.get(record.request_id).state == UNKNOWN for record in sticky)
+
+
 def test_unknown_jobs_tolerates_a_row_resolved_between_pages(tmp_path) -> None:
     """Discovery is read-only and may overlap an inquiry from another worker.
 
@@ -4852,15 +4931,16 @@ def test_a_failure_after_the_call_still_parks_as_unknown(tmp_path) -> None:
     record = queue.submit(_request(key="after-call", source="opencode"))
     claimer = DurableWriteDrain(queue, execute=_never_execute, owner="after-call-owner")
     assert claimer.claim(limit=1) == (record.job_id,)
+    recorder = _Recorder(
+        error=HyMemoryBridgeError("hy_memory_write_verification_authority_invalid"),
+        invoke_hy=True,
+    )
 
     asyncio.run(
         execute_job(
             record.job_id,
             queue=queue,
-            supervisor=_Recorder(
+            supervisor=recorder,
-                error=HyMemoryBridgeError("hy_memory_write_verification_authority_invalid"),
-                invoke_hy=True,
-            ),
             evidence=_EvidenceStore({"e1"}),
             owner="after-call-owner",
         )
@@ -4874,6 +4954,16 @@ def test_a_failure_after_the_call_still_parks_as_unknown(tmp_path) -> None:
 
     assert state == UNKNOWN
     assert invoked == 1
+    assert len(recorder.seen) == 1
+    assert (
+        queue.retry_first_unknown_after_reconcile(
+            older_than=datetime.now(UTC) + timedelta(minutes=1)
+        )
+        == 0
+    )
+    assert queue.retry_unknown_confirmed(record.job_id) == 0
+    assert DurableWriteDrain(queue, execute=_never_execute).claim(limit=1) == ()
+    assert len(recorder.seen) == 1
 
 
 def test_event_time_not_enqueue_order_drives_the_window(

memhub/apps/core/tests/test_memory_tasks.py

@@ -1113,6 +1113,109 @@ def test_task_only_unknown_is_automatically_retried_only_once(tmp_path) -> None:
     assert store.require(task.task_id).status is MemoryTaskStatus.UNKNOWN
 
 
+def test_authority_invalid_task_is_never_reinvoked(tmp_path) -> None:
+    store, profile = _profile_and_tasks(tmp_path)
+    task = store.create(
+        request_id="authority-invalid",
+        user_id="local",
+        kind=MemoryTaskKind.EXTRACT,
+        input_ref="authority-evidence",
+        profile=profile,
+    )
+    with sqlite3.connect(store.database_path) as connection, connection:
+        connection.execute(
+            "UPDATE memory_tasks SET status=?,attempts=1,last_error_code=?,updated_at=? "
+            "WHERE task_id=?",
+            (
+                MemoryTaskStatus.UNKNOWN,
+                "hy_memory_write_verification_authority_invalid",
+                "2026-01-01T00:00:00+00:00",
+                task.task_id,
+            ),
+        )
+
+    cutoff = datetime(2026, 1, 1, 0, 5, tzinfo=UTC)
+    assert store.retry_first_unknown_after_reconcile(older_than=cutoff) == 0
+    assert store.retry_unknown_confirmed() == 0
+    retained = store.require(task.task_id)
+    assert retained.status is MemoryTaskStatus.UNKNOWN
+    assert retained.attempts == 1
+    assert retained.last_error_code == "hy_memory_write_verification_authority_invalid"
+
+
+def test_linked_unknown_is_retried_only_by_the_durable_job(tmp_path) -> None:
+    from one_memory.memory_server.contracts import BatchIdentity, WriteRequest
+    from one_memory.memory_server.durable_queue import DurableWriteQueue
+
+    store, profile = _profile_and_tasks(tmp_path)
+    evidence = EvidenceStore(store.database_path)
+    record = evidence.put(
+        EvidenceInput(
+            source_id="cursor",
+            conversation_id="linked-conversation",
+            message_id="linked-message",
+            role="user",
+            written_by=WrittenBy.PERSON,
+            verbatim="linked unknown",
+            created_at=datetime(2026, 1, 1, tzinfo=UTC),
+            origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+        )
+    ).record
+    task = store.create(
+        request_id="linked-task",
+        user_id="local",
+        kind=MemoryTaskKind.EXTRACT,
+        input_ref=record.evidence_id,
+        profile=profile,
+        input_records=(record,),
+    )
+    queue = DurableWriteQueue(store.database_path, evidence=evidence)
+    job = queue.submit(
+        WriteRequest(
+            internal_session_id="linked-session",
+            batch_identity=BatchIdentity(
+                internal_session_id="linked-session",
+                first_round=1,
+                last_round=1,
+            ),
+            evidence_ids=[record.evidence_id],
+            idempotency_key="linked-job",
+            source_id="cursor",
+        )
+    )
+    with sqlite3.connect(store.database_path) as connection, connection:
+        connection.execute(
+            "UPDATE memory_tasks SET status=?,attempts=1,last_error_code=?,updated_at=? "
+            "WHERE task_id=?",
+            (
+                MemoryTaskStatus.UNKNOWN,
+                "model_content_outcome_unknown",
+                "2026-01-01T00:00:00+00:00",
+                task.task_id,
+            ),
+        )
+        connection.execute(
+            "UPDATE memory_write_jobs SET state='unknown',attempts=1,hy_invoked=1,"
+            "last_error='model_content_outcome_unknown',updated_at=? WHERE job_id=?",
+            ("2026-01-01T00:00:00+00:00", job.job_id),
+        )
+
+    cutoff = datetime(2026, 1, 1, 0, 5, tzinfo=UTC)
+    assert store.retry_first_unknown_after_reconcile(older_than=cutoff) == 0
+    assert store.require(task.task_id).status is MemoryTaskStatus.UNKNOWN
+    assert queue.retry_first_unknown_after_reconcile(older_than=cutoff) == 1
+    assert store.require(task.task_id).status is MemoryTaskStatus.PENDING
+    assert queue.get(job.request_id).state == "pending"
+    assert (
+        store.active_window(
+            now=datetime(2026, 1, 2, tzinfo=UTC),
+            fresh_window=timedelta(days=1),
+            extraction_window=None,
+        )
+        == []
+    )
+
+
 def test_orphan_recovery_preserves_active_running_tasks(tmp_path) -> None:
     store, profile = _profile_and_tasks(tmp_path)
     active = store.create(

memhub/apps/core/tests/test_memory_visibility.py

@@ -8,7 +8,14 @@ import pytest
 from one_memory.imports.grants import GRANT_SCHEMA
 from one_memory.imports.models import EvidenceInput, EvidenceOrigin, OriginMetadata, WrittenBy
 from one_memory.imports.storage import EvidenceStore
-from one_memory.memory_governance import MemoryDraft, MemoryGovernanceStore, MemoryKind
+from one_memory.memory_governance import (
+    HyCreate,
+    HyLink,
+    HyRevise,
+    MemoryDraft,
+    MemoryGovernanceStore,
+    MemoryKind,
+)
 from one_memory.memory_tasks import (
     MemoryTaskOutput,
     MemoryTaskStore,
@@ -19,6 +26,7 @@ from one_memory.memory_visibility import (
     install_triggers,
     pending,
     recompute,
+    startup,
 )
 from one_memory.model_profiles import (
     ModelCapabilities,
@@ -336,6 +344,197 @@ def test_starting_the_app_creates_the_derived_tables(tmp_path):
     asyncio.run(scenario())
 
 
+@pytest.mark.parametrize("visibility_mode", ["exact", "derived"])
+def test_hy_revise_and_repeated_link_reuse_the_same_evidence(library, monkeypatch, visibility_mode):
+    """The outer evidence UPSERT must not override the trigger's duplicate handling."""
+    database, evidence, _tasks, governance, _profile = library
+    monkeypatch.setenv("ONE_MEMORY_READ_VISIBILITY", visibility_mode)
+    startup(database)
+    record = evidence.put(
+        EvidenceInput(
+            source_id="cursor",
+            conversation_id="trigger-regression",
+            message_id="trigger-regression",
+            correlation_id="trigger-regression",
+            role="user",
+            written_by=WrittenBy.PERSON,
+            verbatim="用户住在杭州。",
+            origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+        )
+    ).record
+    created = governance.write_from_hy(
+        HyCreate(
+            draft=MemoryDraft(
+                user_id="local",
+                kind=MemoryKind.CHAT,
+                content="用户住在杭州。",
+                scopes=("all-apps",),
+                evidence_ids=(record.evidence_id,),
+            ),
+            request_id="trigger-create",
+            memory_id="trigger-memory",
+        ),
+        user_id="local",
+        derivation=governance.hy_derivation((record.evidence_id,)),
+    )
+    revision = HyRevise(
+        memory_id=created.memory_id,
+        request_id="trigger-revise",
+        expected_revision=created.revision,
+        content="用户住在杭州并在那里工作。",
+        evidence_ids=(record.evidence_id,),
+    )
+    revised = governance.write_from_hy(
+        revision,
+        user_id="local",
+        derivation=governance.hy_derivation((record.evidence_id,)),
+    )
+    replayed = governance.write_from_hy(
+        revision,
+        user_id="local",
+        derivation=governance.hy_derivation((record.evidence_id,)),
+    )
+    assert replayed == revised
+    link = HyLink(memory_id=created.memory_id, evidence_id=record.evidence_id)
+    for _ in range(2):
+        governance.write_from_hy(
+            link,
+            user_id="local",
+            derivation=governance.hy_derivation((record.evidence_id,)),
+        )
+
+    current = governance.get(created.memory_id, user_id="local")
+    assert current is not None
+    assert current.content == "用户住在杭州并在那里工作。"
+    assert current.revision == created.revision + 1
+    with main_database_read(database) as db:
+        assert db.execute(
+            "SELECT memory_id, COUNT(*) FROM memory_visibility_dirty GROUP BY memory_id"
+        ).fetchall() == [(created.memory_id, 1)]
+
+    if visibility_mode == "derived":
+        monkeypatch.setenv("ONE_MEMORY_READ_VISIBILITY", "exact")
+        exact = {memory.memory_id for memory in SourceReadAccess(database).filter([current])}
+        with main_database_write(database) as db:
+            assert recompute(db, [created.memory_id], readable_sql=_MEMORY_READABLE) == 1
+            db.commit()
+        monkeypatch.setenv("ONE_MEMORY_READ_VISIBILITY", "derived")
+        derived = {memory.memory_id for memory in SourceReadAccess(database).filter([current])}
+        assert derived == exact
+        with main_database_read(database) as db:
+            assert created.memory_id not in pending(db)
+
+
+def test_startup_atomically_replaces_the_legacy_link_update_trigger(tmp_path, monkeypatch):
+    import one_memory.memory_visibility as visibility_module
+    from one_memory.schema_gate import guard_schema_version
+
+    database = tmp_path / "one-memory.sqlite3"
+    expected_version = guard_schema_version(database)
+    evidence = EvidenceStore(database)
+    governance = MemoryGovernanceStore(database)
+    record = evidence.put(
+        EvidenceInput(
+            source_id="cursor",
+            conversation_id="legacy-trigger",
+            message_id="legacy-trigger",
+            correlation_id="legacy-trigger",
+            role="user",
+            written_by=WrittenBy.PERSON,
+            verbatim="升级失败不能改变这条原文。",
+            origin_meta={EvidenceOrigin.FILE: OriginMetadata()},
+        )
+    ).record
+    memory = as_person(governance).create(
+        MemoryDraft(
+            user_id="local",
+            kind=MemoryKind.CHAT,
+            content="升级失败不能改变这条记忆。",
+            scopes=("all-apps",),
+            evidence_ids=(record.evidence_id,),
+        ),
+        request_id="legacy-trigger-memory",
+    )
+    startup(database)
+    with sqlite3.connect(database) as db:
+        db.execute("DROP TRIGGER memory_visibility_dirty_on_link_update")
+        db.executescript(
+            """
+            CREATE TRIGGER memory_visibility_dirty_on_link_update
+            AFTER UPDATE ON governed_memory_evidence
+            BEGIN
+                INSERT OR IGNORE INTO memory_visibility_dirty (memory_id)
+                VALUES (NEW.memory_id);
+                INSERT OR IGNORE INTO memory_visibility_dirty (memory_id)
+                VALUES (OLD.memory_id);
+            END;
+            """
+        )
+        db.execute(
+            "INSERT INTO memory_visibility (memory_id, visible, reason, computed_at) "
+            "VALUES (?, 1, 'ok', '2026-10-19T00:00:00+00:00')",
+            (memory.memory_id,),
+        )
+        db.execute(
+            "INSERT INTO memory_visibility_dirty (memory_id) VALUES (?)",
+            (memory.memory_id,),
+        )
+        legacy_trigger = db.execute(
+            "SELECT sql FROM sqlite_master "
+            "WHERE type='trigger' AND name='memory_visibility_dirty_on_link_update'"
+        ).fetchone()[0]
+        before = tuple(db.iterdump())
+
+    real_trigger_sql = visibility_module._trigger_sql
+
+    def fail_after_drop() -> list[tuple[str, str]]:
+        statements = real_trigger_sql()
+        return [
+            (table, "CREATE TRIGGER")
+            if "CREATE TRIGGER IF NOT EXISTS memory_visibility_dirty_on_link_update" in statement
+            else (table, statement)
+            for table, statement in statements
+        ]
+
+    with monkeypatch.context() as failure:
+        failure.setattr(visibility_module, "_trigger_sql", fail_after_drop)
+        with pytest.raises(sqlite3.OperationalError):
+            startup(database)
+
+    with sqlite3.connect(database) as db:
+        assert tuple(db.iterdump()) == before
+        assert (
+            db.execute(
+                "SELECT sql FROM sqlite_master "
+                "WHERE type='trigger' AND name='memory_visibility_dirty_on_link_update'"
+            ).fetchone()[0]
+            == legacy_trigger
+        )
+        assert db.execute("SELECT MAX(version) FROM schema_migrations").fetchone()[0] == (
+            expected_version
+        )
+
+    startup(database)
+    startup(database)
+
+    with sqlite3.connect(database) as db:
+        trigger_sql = db.execute(
+            "SELECT sql FROM sqlite_master "
+            "WHERE type='trigger' AND name='memory_visibility_dirty_on_link_update'"
+        ).fetchone()[0]
+        assert trigger_sql.count("ON CONFLICT(memory_id) DO NOTHING") == 2
+        assert "INSERT OR IGNORE" not in trigger_sql
+        assert db.execute("SELECT MAX(version) FROM schema_migrations").fetchone()[0] == (
+            expected_version
+        )
+        assert db.execute("SELECT * FROM memory_visibility").fetchall() == [
+            (memory.memory_id, 1, "ok", "2026-10-19T00:00:00+00:00")
+        ]
+        assert db.execute("SELECT * FROM memory_visibility_dirty").fetchall() == [
+            (memory.memory_id,)
+        ]
+
+
 def test_the_audit_notices_a_conclusion_that_no_longer_holds(library):
     """The ongoing proof: recompute a slice and compare with what is stored.
 

memhub/apps/core/tests/test_model_profile_api.py

@@ -259,12 +259,17 @@ def test_activation_survives_a_corpus_already_imported(tmp_path) -> None:
                 ),
                 secrets=secrets,
             )
+            readmissions: list[bool] = []
+            app.state.memory_write_queue.readmit_model_profile_waiters = (
+                lambda: readmissions.append(True) or 0
+            )
 
             activated = await client.post(f"/api/v1/model-profiles/{profile_id}/activate")
 
             assert activated.status_code == 200
             assert activated.json()["is_active"] is True
             assert API_KEY not in activated.text
+            assert readmissions == [True]
             # Started, not waited for: the walk reads every conversation.
             assert not app.state.memory_backfill.done()
 

memhub/apps/core/tests/test_policy_api_wiring.py

@@ -111,6 +111,9 @@ def test_the_manual_source_is_admitted_by_grant_without_a_toggle(tmp_path) -> No
     app = SimpleNamespace(
         state=SimpleNamespace(
             config_store=config,
+            model_profile_store=SimpleNamespace(
+                active_identity=lambda: ("profile", 1)
+            ),
             source_grant_store=SimpleNamespace(
                 receipt=lambda **_kwargs: SimpleNamespace(revoked_at=None)
             ),
@@ -125,6 +128,21 @@ def test_the_manual_source_is_admitted_by_grant_without_a_toggle(tmp_path) -> No
     assert _sedimentation_claim_enabled(app, "one-memory-manual") is False
 
 
+def test_claim_gate_waits_without_polling_when_no_model_is_active(tmp_path) -> None:
+    config = ConfigStore(tmp_path)
+    app = SimpleNamespace(
+        state=SimpleNamespace(
+            config_store=config,
+            model_profile_store=SimpleNamespace(active_identity=lambda: None),
+            source_grant_store=SimpleNamespace(
+                receipt=lambda **_kwargs: SimpleNamespace(revoked_at=None)
+            ),
+        )
+    )
+
+    assert _sedimentation_claim_enabled(app, "one-memory-manual") is False
+
+
 def test_first_import_gate_waits_for_every_selected_history_job() -> None:
     state = SimpleNamespace(
         status="importing",

memhub/apps/web/src/FirstImportReview.test.tsx

@@ -174,20 +174,20 @@ it("continues to step3 while sample GET is pending without calling completion AP
   expect(screen.queryByText(/^已生成.*记忆/)).toBeNull();
 });
 it.each([
-  "pending",
+  ["pending", "已扫描 100 条，正在筛选"],
-  "paused",
+  ["paused", "已扫描 100 条，尚未保存"],
-  "interrupted",
+  ["interrupted", "已扫描 100 条，尚未保存"],
-  "failed",
+  ["failed", "已扫描 100 条，尚未保存"],
-  "cancelled",
+  ["cancelled", "已扫描 100 条，尚未保存"],
-  "completed",
+  ["completed", "已扫描 100 条，本次没有可保存的对话"],
 ] as const)(
   "shows %s without mutating jobs or fetching missing content",
-  (status) => {
+  (status, expected) => {
     setup({
       jobs: [{ ...job, status, stored: 0 }],
       sources: [{ ...source, imported_messages: 0 }],
     });
-    expect(screen.getByText("已扫描 100 条，正在保存")).toBeTruthy();
+    expect(screen.getByText(expected)).toBeTruthy();
     expect(api.receipt).not.toHaveBeenCalled();
   },
 );

memhub/apps/web/src/R6ImportProgress.test.tsx

@@ -38,6 +38,71 @@ it("uses the selected job's message count, never the source total or fake percen
   expect(container.textContent).not.toContain("%");
   expect(container.textContent).not.toContain("持续导入中");
 });
+
+it("does not describe scanned-but-skipped history as saved", () => {
+  const { container } = render(
+    <R6ImportProgress
+      sourceIds={["cursor"]}
+      sources={[source]}
+      jobs={[
+        {
+          ...job,
+          status: "completed",
+          discovered: 12_375,
+          stored: 0,
+          skipped: 12_375,
+          history_after: "2026-10-15T00:00:00Z",
+        },
+      ]}
+    />,
+  );
+  expect(
+    screen.getByText("已扫描 12,375 条，本次没有符合时间范围的对话"),
+  ).toBeTruthy();
+  expect(screen.getByText("导入完成")).toBeTruthy();
+  expect(container.textContent).not.toContain("正在保存");
+  expect(container.textContent).not.toContain("已保存 12,375");
+});
+
+it("calls an active bounded scan filtering rather than saving", () => {
+  render(
+    <R6ImportProgress
+      sourceIds={["cursor"]}
+      sources={[source]}
+      jobs={[
+        {
+          ...job,
+          discovered: 3_522,
+          stored: 0,
+          history_after: "2026-10-15T00:00:00Z",
+        },
+      ]}
+    />,
+  );
+  expect(screen.getByText("已扫描 3,522 条，正在筛选最近 3 天")).toBeTruthy();
+});
+
+it("reports only stored rows as saved after a completed partial import", () => {
+  const { container } = render(
+    <R6ImportProgress
+      sourceIds={["cursor"]}
+      sources={[source]}
+      jobs={[
+        {
+          ...job,
+          status: "completed",
+          discovered: 25_177,
+          stored: 20_826,
+          skipped: 4_351,
+        },
+      ]}
+    />,
+  );
+  expect(screen.getByText("已保存 20,826 条消息")).toBeTruthy();
+  expect(screen.getByText("导入完成")).toBeTruthy();
+  expect(container.textContent).not.toContain("25,177");
+});
+
 it.each(["paused", "interrupted", "failed", "completed", "cancelled"] as const)(
   "does not animate %s as active work",
   (status) => {

memhub/apps/web/src/R6ImportProgress.tsx

@@ -3,6 +3,21 @@ import { useEffect, useLayoutEffect, useRef } from "react";
 import { AppMark } from "./app-mark";
 import "./r6-import-progress.css";
 
+function progressText(job: ImportJob): string | null {
+  if (job.stored > 0) return `已保存 ${job.stored.toLocaleString()} 条消息`;
+  if (job.discovered <= 0) return null;
+  const scanned = `已扫描 ${job.discovered.toLocaleString()} 条`;
+  if (job.status === "completed")
+    return job.history_after
+      ? `${scanned}，本次没有符合时间范围的对话`
+      : `${scanned}，本次没有可保存的对话`;
+  if (job.status === "pending" || job.status === "running")
+    return job.history_after
+      ? `${scanned}，正在筛选最近 3 天`
+      : `${scanned}，正在筛选`;
+  return `${scanned}，尚未保存`;
+}
+
 export function R6ImportProgress({
   sourceIds,
   sources,
@@ -86,8 +101,9 @@ export function R6ImportProgress({
           interrupted: "已中断",
           failed: "导入遇到问题",
           cancelled: "已取消",
-          completed: "已保存",
+          completed: "导入完成",
         };
+        const progress = job ? progressText(job) : null;
         return (
           <article
             className="r6-import-progress-card"
@@ -102,10 +118,8 @@ export function R6ImportProgress({
               </div>
             </div>
             <h2>{source?.display_name ?? id}</h2>
-            {job && job.stored > 0 ? (
+            {progress ? (
-              <p>{`已保存 ${job.stored.toLocaleString()} 条消息`}</p>
+              <p>{progress}</p>
-            ) : job && job.discovered > 0 ? (
-              <p>{`已扫描 ${job.discovered.toLocaleString()} 条，正在保存`}</p>
             ) : noImportedHistory ? (
               <p>没有可保存的历史对话</p>
             ) : null}

memhub/packages/runtime-control/src/release-download.test.ts

@@ -312,6 +312,48 @@ describe("remote Runtime acquisition", () => {
     });
   });
 
+  it("allows the exact GitHub Release asset redirect hosts by default", async () => {
+    const value = await fixture();
+    const manifestUrl =
+      "https://github.com/example/project/releases/download/v1.2.3/release-manifest.json";
+    const artifactUrl =
+      "https://github.com/example/project/releases/download/v1.2.3/runtime.tar.gz";
+    const manifestAssetUrl =
+      "https://release-assets.githubusercontent.com/manifest-download";
+    const runtimeAssetUrl =
+      "https://release-assets.githubusercontent.com/runtime-download";
+    const release = {
+      ...value.release,
+      artifacts: [{ ...value.artifact, url: artifactUrl }],
+    };
+    const temporaryParent = await mkdtemp(
+      path.join(os.tmpdir(), "memhub-github-release-parent-"),
+    );
+    roots.push(temporaryParent);
+
+    const acquired = await acquireRuntimeDistribution({
+      manifestUrl,
+      expectedVersion: "1.2.3",
+      channel: "stable",
+      platform: "linux",
+      arch: "x64",
+      temporaryParent,
+      httpGet: httpFixture({
+        [manifestUrl]: response(new Uint8Array(), 302, {
+          location: manifestAssetUrl,
+        }),
+        [manifestAssetUrl]: response(Buffer.from(JSON.stringify(release))),
+        [artifactUrl]: response(new Uint8Array(), 302, {
+          location: runtimeAssetUrl,
+        }),
+        [runtimeAssetUrl]: response(value.archiveBytes),
+      }),
+    });
+
+    expect(acquired.release.version).toBe("1.2.3");
+    await acquired.cleanup();
+  });
+
   it("rejects an archive whose internal commit drifts from the Release", async () => {
     const value = await fixture();
     const release = { ...value.release, git_commit: "b".repeat(40) };
@@ -368,7 +410,6 @@ describe("remote Runtime acquisition", () => {
   );
 
   it("rejects a redirect to an untrusted host", async () => {
-    const value = await fixture();
     await expect(
       acquireRuntimeDistribution({
         manifestUrl: "https://downloads.example.test/release-manifest.json",

memhub/packages/runtime-control/src/release-download.ts

@@ -38,6 +38,10 @@ const SHA256 = /^[a-f0-9]{64}$/;
 const COMMIT = /^[0-9a-f]{40}$/;
 const SAFE_FILE_NAME = /^[0-9A-Za-z][0-9A-Za-z._-]{0,255}$/;
 const REDIRECTS = new Set([301, 302, 303, 307, 308]);
+const GITHUB_RELEASE_REDIRECT_HOSTS = [
+  "release-assets.githubusercontent.com",
+  "objects.githubusercontent.com",
+] as const;
 
 export type ReleaseChannel = "stable" | "prerelease";
 export type RuntimeReleasePlatform = "darwin" | "linux";
@@ -330,8 +334,12 @@ export async function resolveRuntimeRelease(options: {
   maxRedirects?: number;
 }): Promise<ResolvedRuntimeRelease> {
   const manifestUrl = trustedUrl(options.manifestUrl);
+  const defaultHosts =
+    normalizedHost(manifestUrl.hostname) === "github.com"
+      ? [manifestUrl.hostname, ...GITHUB_RELEASE_REDIRECT_HOSTS]
+      : [manifestUrl.hostname];
   const allowedHosts = new Set(
-    (options.allowedHosts ?? [manifestUrl.hostname]).map(normalizedHost),
+    (options.allowedHosts ?? defaultHosts).map(normalizedHost),
   );
   if (!allowedHosts.has(normalizedHost(manifestUrl.hostname))) {
     throw new Error("release_url_untrusted");

memhub/scripts/build-release-assets.mjs

@@ -8,6 +8,7 @@ import {
   mkdir,
   readdir,
   readFile,
+  readlink,
   realpath,
   rm,
   stat,
@@ -320,6 +321,12 @@ async function distributionBytes(source) {
 
 async function sha256File(file) {
   const digest = createHash("sha256");
+  const info = await lstat(file);
+  if (info.isSymbolicLink()) {
+    digest.update("symlink\0");
+    digest.update(await readlink(file));
+    return digest.digest("hex");
+  }
   for await (const chunk of createReadStream(file)) digest.update(chunk);
   return digest.digest("hex");
 }

memhub/scripts/build-release-assets.test.mjs

@@ -6,6 +6,7 @@ import {
   readdir,
   readFile,
   rm,
+  symlink,
   writeFile,
 } from "node:fs/promises";
 import os from "node:os";
@@ -121,6 +122,35 @@ test("builds versioned Runtime archives, checksums and a downloader-compatible m
   }
 });
 
+test("verifies Runtime symlinks with the same identity rule as the installer", async (t) => {
+  const root = await mkdtemp(path.join(os.tmpdir(), "memhub-release-assets-"));
+  t.after(() => rm(root, { recursive: true, force: true }));
+  const source = await distribution(root, "linux", "x64");
+  const link = "memhub";
+  await symlink(link, path.join(source, "bin", "node"));
+  const manifestPath = path.join(source, "manifest.json");
+  const manifest = JSON.parse(await readFile(manifestPath, "utf8"));
+  manifest.files["bin/node"] = sha256(`symlink\0${link}`);
+  await writeFile(manifestPath, `${JSON.stringify(manifest, null, 2)}\n`);
+
+  const release = await buildReleaseAssets({
+    output: path.join(root, "release"),
+    version: "0.1.0-internal.1",
+    channel: "prerelease",
+    assetBaseUrl: "https://git.example.test/release/downloads/",
+    distributions: [source],
+  });
+
+  assert.ok(
+    release.artifacts.some(
+      (artifact) =>
+        artifact.kind === "runtime" &&
+        artifact.platform === "linux" &&
+        artifact.arch === "x64",
+    ),
+  );
+});
+
 test("refuses version, build and duplicate target drift", async (t) => {
   const root = await mkdtemp(path.join(os.tmpdir(), "memhub-release-assets-"));
   t.after(() => rm(root, { recursive: true, force: true }));

memhub/scripts/publish-cli-after-release.mjs

@@ -23,7 +23,8 @@ export async function publishCliAfterRelease(options, dependencies = {}) {
   const statImpl = dependencies.statImpl ?? stat;
   const buildEnvironment = {
     ...environment,
-    ONE_MEMORY_VERSION: options.version,
+    MEMHUB_NPM_VERSION: options.version,
+    MEMHUB_DEFAULT_RUNTIME_VERSION: options.runtimeVersion,
     ONE_MEMORY_BUILD_NUMBER: String(options.buildNumber),
     ONE_MEMORY_COMMIT: options.gitCommit,
   };
@@ -31,7 +32,7 @@ export async function publishCliAfterRelease(options, dependencies = {}) {
   const versionReceipt = await verifyVersion(
     [
       "--version",
-      options.version,
+      options.runtimeVersion,
       "--channel",
       options.channel,
       "--tag",
@@ -43,7 +44,7 @@ export async function publishCliAfterRelease(options, dependencies = {}) {
   );
   const releaseReceipt = await verifyRelease({
     manifestUrl: options.manifestUrl,
-    version: options.version,
+    version: options.runtimeVersion,
     gitCommit: options.gitCommit,
     channel: options.channel,
     platform: options.platform,
@@ -106,7 +107,8 @@ export async function publishCliAfterRelease(options, dependencies = {}) {
   return {
     schema_version: 1,
     product: "memhub-cli-publication",
-    version: versionReceipt.version,
+    version: options.version,
+    runtime_version: releaseReceipt.version,
     build_number: versionReceipt.build_number,
     tag: versionReceipt.tag,
     release: releaseReceipt,
@@ -196,6 +198,11 @@ function parseArgs(arguments_) {
       options.manifestUrl = value;
     else if (flag === "--version" && options.version === undefined)
       options.version = value;
+    else if (
+      flag === "--runtime-version" &&
+      options.runtimeVersion === undefined
+    )
+      options.runtimeVersion = value;
     else if (flag === "--git-commit" && options.gitCommit === undefined)
       options.gitCommit = value;
     else if (flag === "--channel" && options.channel === undefined)
@@ -219,6 +226,7 @@ function parseArgs(arguments_) {
   if (
     !options.manifestUrl ||
     !options.version ||
+    !options.runtimeVersion ||
     !options.gitCommit ||
     !options.channel ||
     !options.tag ||
@@ -230,7 +238,7 @@ function parseArgs(arguments_) {
     !options.npmTag
   ) {
     throw new Error(
-      "usage: publish-cli-after-release.mjs --manifest-url <https-url> --version <SemVer> --git-commit <sha> --channel <stable|prerelease> --tag <vSemVer> --build-number <integer> --platform <darwin|linux> --arch <x64|arm64> --registry <https-url> --npm-tag <tag> [--access <public|restricted>] [--provenance] [--allowed-host <hostname>] [--dry-run]",
+      "usage: publish-cli-after-release.mjs --manifest-url <https-url> --version <CLI SemVer> --runtime-version <Runtime SemVer> --git-commit <sha> --channel <stable|prerelease> --tag <vSemVer> --build-number <integer> --platform <darwin|linux> --arch <x64|arm64> --registry <https-url> --npm-tag <tag> [--access <public|restricted>] [--provenance] [--allowed-host <hostname>] [--dry-run]",
     );
   }
   return options;

memhub/scripts/publish-cli-after-release.test.mjs

@@ -6,6 +6,9 @@ import test from "node:test";
 
 import { publishCliAfterRelease } from "./publish-cli-after-release.mjs";
 
+const CLI_VERSION = "0.1.1";
+const RUNTIME_VERSION = "0.1.0-internal.1";
+
 async function fixture(t) {
   const root = await mkdtemp(path.join(os.tmpdir(), "memhub-publish-gate-"));
   t.after(() => rm(root, { recursive: true, force: true }));
@@ -15,10 +18,10 @@ async function fixture(t) {
   await mkdir(output, { recursive: true });
   await writeFile(
     path.join(publish, "package.json"),
-    `${JSON.stringify({ name: "@example/memhub-cli", version: "0.1.0-internal.1", private: false })}\n`,
+    `${JSON.stringify({ name: "@example/memhub-cli", version: CLI_VERSION, private: false })}\n`,
   );
   await writeFile(
-    path.join(output, "example-memhub-cli-0.1.0-internal.1.tgz"),
+    path.join(output, `example-memhub-cli-${CLI_VERSION}.tgz`),
     "tarball",
   );
   return root;
@@ -26,10 +29,11 @@ async function fixture(t) {
 
 const options = {
   manifestUrl: "https://releases.example.test/release-manifest.json",
-  version: "0.1.0-internal.1",
+  version: CLI_VERSION,
+  runtimeVersion: RUNTIME_VERSION,
   gitCommit: "a".repeat(40),
   channel: "prerelease",
-  tag: "v0.1.0-internal.1",
+  tag: `v${RUNTIME_VERSION}`,
   buildNumber: 1,
   platform: "linux",
   arch: "x64",
@@ -49,15 +53,24 @@ test("verifies the Runtime before building and publishing the CLI", async (t) =>
     {
       verifyVersion: async () => {
         events.push("version");
-        return { version: options.version, build_number: 1, tag: options.tag };
+        return {
+          version: options.runtimeVersion,
+          build_number: 1,
+          tag: options.tag,
+        };
       },
       verifyRelease: async (verifyOptions) => {
         events.push("release");
         assert.equal(verifyOptions.gitCommit, options.gitCommit);
-        return { version: options.version, sha256: "a".repeat(64) };
+        assert.equal(verifyOptions.version, options.runtimeVersion);
+        return { version: options.runtimeVersion, sha256: "a".repeat(64) };
       },
       runCommand: (command, arguments_, commandOptions) => {
-        assert.equal(commandOptions.env.ONE_MEMORY_VERSION, options.version);
+        assert.equal(commandOptions.env.MEMHUB_NPM_VERSION, options.version);
+        assert.equal(
+          commandOptions.env.MEMHUB_DEFAULT_RUNTIME_VERSION,
+          options.runtimeVersion,
+        );
         assert.equal(
           commandOptions.env.ONE_MEMORY_BUILD_NUMBER,
           String(options.buildNumber),
@@ -74,6 +87,8 @@ test("verifies the Runtime before building and publishing the CLI", async (t) =>
   assert.match(events[3], /--access public/);
   assert.match(events[3], /--provenance/);
   assert.equal(receipt.npm_package, "@example/memhub-cli");
+  assert.equal(receipt.version, CLI_VERSION);
+  assert.equal(receipt.runtime_version, RUNTIME_VERSION);
   assert.equal(receipt.dry_run, true);
   assert.equal(receipt.npm_access, "public");
   assert.equal(receipt.provenance, true);
@@ -87,7 +102,7 @@ test("never builds or invokes npm when the Runtime release is unavailable", asyn
       { ...options, root },
       {
         verifyVersion: async () => ({
-          version: options.version,
+          version: options.runtimeVersion,
           build_number: 1,
           tag: options.tag,
         }),
@@ -115,11 +130,11 @@ test("rejects a package identity that differs from the configured publish name",
       },
       {
         verifyVersion: async () => ({
-          version: options.version,
+          version: options.runtimeVersion,
           build_number: 1,
           tag: options.tag,
         }),
-        verifyRelease: async () => ({ version: options.version }),
+        verifyRelease: async () => ({ version: options.runtimeVersion }),
         runCommand: () => undefined,
       },
     ),

tests/coding/test_fact_reconciler.py

@@ -107,6 +107,7 @@ class TestSupersede:
     async def _seed_old(self, store, *, derived=None):
         old = WorkingFact(
             fact_id="f_old", user_id="u1",
+            session_id="test-session",
             statement=(
                 "用户要修复 CI 的 npm run build；检查 .github/workflows/ci.yml 后"
                 "定位到 Node 18 报 ERR_OSSL_EVP_UNSUPPORTED，尚未修改。"
@@ -129,7 +130,7 @@ class TestSupersede:
             [_draft(
                 "用户继续修复同一 CI 构建；把 .github/workflows/ci.yml 升到 Node 20，"
                 "npm run build 已通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         new_id = facts[0].fact_id
@@ -152,7 +153,7 @@ class TestSupersede:
             [_draft(
                 "用户继续修复同一 CI 构建；把 .github/workflows/ci.yml 升到 Node 20，"
                 "npm run build 已通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
         new_id = facts[0].fact_id
 
@@ -184,7 +185,9 @@ class TestSupersede:
                 coding_store, mock_embed,
                 _llm([{"draft_idx": 0, "supersedes": [target or prev_id]}]),
             )
-            facts, _ = await rec.reconcile([_draft(stmt)], user_id="u1")
+            facts, _ = await rec.reconcile(
+                [_draft(stmt)], user_id="u1", session_id="test-session"
+            )
             prev_id = facts[0].fact_id
 
         lineage = await coding_store.get_fact_ancestors(prev_id, user_id="u1")
@@ -208,6 +211,7 @@ class TestSupersede:
                 "配置完成但尚未发布。"
             )],
             user_id="u1",
+            session_id="test-session",
         )
 
         prompt = llm.calls[0]["prompt"]
@@ -234,7 +238,9 @@ class TestSupersede:
             coding_store, mock_embed, _llm([{"draft_idx": 0, "supersedes": ["f_old"]}]),
         )
         await rec.reconcile(
-            [_draft("config 位于 /etc/app/config.yaml")], user_id="u1",
+            [_draft("config 位于 /etc/app/config.yaml")],
+            user_id="u1",
+            session_id="test-session",
         )
 
         # 拿旧说法原文去搜，只能落到新事实（靠两者本身相似），
@@ -256,6 +262,34 @@ class TestSupersede:
         old = await coding_store.get_fact("f_old", user_id="u1")
         assert old.status == "active"
 
+    async def test_near_duplicate_fact_cannot_cross_authority_scope(
+        self, coding_store, mock_embed
+    ):
+        await coding_store.insert_fact(WorkingFact(
+            fact_id="f_old",
+            user_id="u1",
+            agent_id="scope:2026-10-18",
+            session_id="conversation-a",
+            workspace_id="repo-a",
+            statement="CI build now passes after upgrading Node",
+        ))
+        llm = _llm([{"draft_idx": 0, "supersedes": ["f_old"]}])
+        rec = WorkingFactReconciler(coding_store, mock_embed, llm)
+
+        _, ops = await rec.reconcile(
+            [_draft("CI build now passes after upgrading Node")],
+            user_id="u1",
+            agent_id="scope:2026-10-19",
+            session_id="conversation-b",
+            workspace_id="repo-a",
+        )
+
+        old = await coding_store.get_fact("f_old", user_id="u1")
+        assert old is not None
+        assert old.status == "active"
+        assert [op.action for op in ops] == ["INSERT"]
+        assert len(llm.calls) == 0
+
     async def test_old_fact_superseded_only_once(self, coding_store, mock_embed):
         """两条新事实都声称覆盖同一条旧事实时，只让第一条生效。
 
@@ -274,6 +308,7 @@ class TestSupersede:
                 _draft("config 位于 /srv/app/config.yaml"),
             ],
             user_id="u1",
+            session_id="test-session",
         )
         assert len(facts) == 2
         assert sum(1 for o in ops if o.action == "SUPERSEDE") == 1
@@ -288,6 +323,7 @@ class TestChainSummary:
     async def _seed_old(self, store):
         await store.insert_fact(WorkingFact(
             fact_id="f_old", user_id="u1",
+            session_id="test-session",
             statement=(
                 "用户要迁移 app config；确认当前读取 /opt/app/config.yaml，"
                 "尚未修改配置加载逻辑。"
@@ -310,7 +346,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         stored = await coding_store.get_fact(facts[0].fact_id, user_id="u1")
@@ -326,7 +362,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         assert len(llm.calls) == 2, "SUPERSEDE 应触发第二次调用"
@@ -344,7 +380,11 @@ class TestChainSummary:
         llm = _llm([])
         rec = WorkingFactReconciler(coding_store, mock_embed, llm)
 
-        await rec.reconcile([_draft("另一个话题的结论")], user_id="u1")
+        await rec.reconcile(
+            [_draft("另一个话题的结论")],
+            user_id="u1",
+            session_id="test-session",
+        )
         assert len(llm.calls) == 1
 
     async def test_prompt_carries_full_lineage(self, coding_store, mock_embed):
@@ -357,7 +397,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         llm2 = _llm([{"draft_idx": 0, "supersedes": [facts[0].fact_id]}])
@@ -366,7 +406,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续完善同一 config 迁移；改由环境变量 APP_CONFIG 指定路径，"
                 "回归测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         prompt = llm2.calls[1]["prompt"]
@@ -393,7 +433,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         assert sum(1 for o in ops if o.action == "SUPERSEDE") == 1
@@ -426,7 +466,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
         assert facts[0].fact_chain_summary == ""
 
@@ -436,7 +476,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续完善同一 config 迁移；改由环境变量 APP_CONFIG 指定路径，"
                 "回归测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         prompt = llm2.calls[1]["prompt"]
@@ -450,7 +490,8 @@ class TestChainSummary:
         total = MAX_FACT_CHAIN_ENTRIES + 3
         for i in range(total):
             await coding_store.insert_fact(WorkingFact(
-                fact_id=f"g{i}", user_id="u1", statement=f"第 {i} 代结论",
+                fact_id=f"g{i}", user_id="u1", session_id="test-session",
+                statement=f"第 {i} 代结论",
                 created_at=datetime(2026, 1, i + 1),
             ))
             if prev:
@@ -459,7 +500,9 @@ class TestChainSummary:
 
         llm = _llm([{"draft_idx": 0, "supersedes": [prev]}])
         rec = WorkingFactReconciler(coding_store, mock_embed, llm)
-        await rec.reconcile([_draft("最新结论")], user_id="u1")
+        await rec.reconcile(
+            [_draft("最新结论")], user_id="u1", session_id="test-session"
+        )
 
         prompt = llm.calls[1]["prompt"]
         assert "第 0 代结论" not in prompt, "最老的几代该被截掉"
@@ -478,7 +521,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
         entries = llm.calls[1]["prompt"].split(
             "ENTRIES\n═══════════════════════════════════════════════════════════════════\n",
@@ -497,7 +540,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         rec2 = WorkingFactReconciler(
@@ -509,7 +552,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续完善同一 config 迁移；改由环境变量 APP_CONFIG 指定路径，"
                 "回归测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         stored = await coding_store.get_fact(facts2[0].fact_id, user_id="u1")
@@ -526,7 +569,7 @@ class TestChainSummary:
             [_draft(
                 "用户继续迁移 app config；将加载路径改为 /etc/app/config.yaml，"
                 "配置加载测试通过。"
-            )], user_id="u1",
+            )], user_id="u1", session_id="test-session",
         )
 
         assert len(llm.calls) == 1
@@ -564,7 +607,8 @@ class TestFailSafe:
     async def test_llm_failure_still_inserts(self, coding_store, mock_embed):
         """LLM 挂了宁可多留一条过期事实，也不能因为一次调用失败丢结论。"""
         await coding_store.insert_fact(WorkingFact(
-            fact_id="f_old", user_id="u1", statement="config 位于 /opt/app/config.yaml",
+            fact_id="f_old", user_id="u1", session_id="test-session",
+            statement="config 位于 /opt/app/config.yaml",
         ))
         rec = WorkingFactReconciler(coding_store, mock_embed, FailingLLMProvider())
 
@@ -578,7 +622,8 @@ class TestFailSafe:
 
     async def test_unparsable_llm_output(self, coding_store, mock_embed):
         await coding_store.insert_fact(WorkingFact(
-            fact_id="f_old", user_id="u1", statement="config 位于 /opt/app/config.yaml",
+            fact_id="f_old", user_id="u1", session_id="test-session",
+            statement="config 位于 /opt/app/config.yaml",
         ))
         rec = WorkingFactReconciler(
             coding_store, mock_embed, MockLLMProvider("不是 JSON 的一段话")
@@ -592,7 +637,8 @@ class TestFailSafe:
 
     async def test_markdown_fence_parsed(self, coding_store, mock_embed):
         await coding_store.insert_fact(WorkingFact(
-            fact_id="f_old", user_id="u1", statement="config 位于 /opt/app/config.yaml",
+            fact_id="f_old", user_id="u1", session_id="test-session",
+            statement="config 位于 /opt/app/config.yaml",
         ))
         llm = MockLLMProvider([
             "```json\n" + _ops_resp([{"draft_idx": 0, "supersedes": ["f_old"]}]) + "\n```",
@@ -601,7 +647,9 @@ class TestFailSafe:
         rec = WorkingFactReconciler(coding_store, mock_embed, llm)
 
         _, ops = await rec.reconcile(
-            [_draft("config 位于 /etc/app/config.yaml")], user_id="u1",
+            [_draft("config 位于 /etc/app/config.yaml")],
+            user_id="u1",
+            session_id="test-session",
         )
         assert sum(1 for o in ops if o.action == "SUPERSEDE") == 1
 
@@ -618,7 +666,8 @@ class TestSubjectGateSwitch:
 
     async def _seed(self, store, fact_id, subject, statement):
         await store.insert_fact(WorkingFact(
-            fact_id=fact_id, user_id="u1", subject=subject, statement=statement,
+            fact_id=fact_id, user_id="u1", session_id="test-session",
+            subject=subject, statement=statement,
         ))
 
     async def test_gate_on_blocks_cross_subject_supersede(self, coding_store, mock_embed):
@@ -632,6 +681,7 @@ class TestSubjectGateSwitch:
         _, ops = await rec.reconcile(
             [WorkingFactDraft(subject="Bart", statement="Bart 的三日行程已排定，预算待定。")],
             user_id="u1",
+            session_id="test-session",
         )
         assert [o.action for o in ops] == ["INSERT"]
         assert (await coding_store.get_fact("f_emma", user_id="u1")).status == "active"
@@ -647,6 +697,7 @@ class TestSubjectGateSwitch:
         facts, ops = await rec.reconcile(
             [WorkingFactDraft(subject="Bart", statement="Bart 的三日行程已排定，预算待定。")],
             user_id="u1",
+            session_id="test-session",
         )
         assert [o.action for o in ops] == ["SUPERSEDE", "INSERT"]
         old = await coding_store.get_fact("f_emma", user_id="u1")
@@ -667,7 +718,8 @@ class TestSubjectGateSwitch:
             [WorkingFactDraft(subject="Bart", statement="Bart 的行程。")],
             [[old]],
             {0: ["f_emma"]},
-            user_id="u1", agent_id="a1", session_id=None, workspace_id=None,
+            user_id="u1", agent_id="default_agent",
+            session_id="test-session", workspace_id=None,
         )
         assert "SUPERSEDE" in [o.action for o in ops]
 
@@ -687,7 +739,8 @@ class TestSubjectGateSwitch:
                 [WorkingFactDraft(subject="deploy.sh", statement="新的部署脚本结论。")],
                 [[old]],
                 {0: [f"f_noneed_{gate}"]},
-                user_id="u1", agent_id="a1", session_id=None, workspace_id=None,
+                user_id="u1", agent_id="default_agent",
+                session_id="test-session", workspace_id=None,
             )
             assert "SUPERSEDE" in [o.action for o in ops], f"gate={gate} 时空 subject 被误拦"
 
@@ -714,6 +767,7 @@ class TestDetailsInheritance:
     async def test_ancestor_details_copied_verbatim(self, coding_store, mock_embed):
         await coding_store.insert_fact(WorkingFact(
             fact_id="f_old", user_id="u1", subject="Emma",
+            session_id="test-session",
             statement="Emma 的行程草案已出，预算待核。",
             details="日均餐费 $32.4-$57.6\n酒店评分 3.2-3.8",
         ))
@@ -726,6 +780,7 @@ class TestDetailsInheritance:
             [WorkingFactDraft(subject="Emma", statement="Emma 的行程已定稿。",
                               details="航班 CA1234")],
             user_id="u1",
+            session_id="test-session",
         )
         stored = await coding_store.get_fact(facts[0].fact_id, user_id="u1")
         # 本代的 details 在，祖先的具体取值也一字不差地在
@@ -742,6 +797,7 @@ class TestDetailsInheritance:
         """
         await coding_store.insert_fact(WorkingFact(
             fact_id="f_g1", user_id="u1", subject="T",
+            session_id="test-session",
             statement="第一代结论。", details="V1=" + "a" * 300,
         ))
         prev = "f_g1"
@@ -756,6 +812,7 @@ class TestDetailsInheritance:
                 [WorkingFactDraft(subject="T", statement=f"第{gen}代结论。",
                                   details=f"V{gen}=" + "a" * 300)],
                 user_id="u1",
+                session_id="test-session",
             )
             prev = facts[0].fact_id
             stored = await coding_store.get_fact(prev, user_id="u1")
@@ -772,6 +829,7 @@ class TestDetailsInheritance:
         """预算是硬的，但被丢掉的代数必须写出来，不静默截断。"""
         await coding_store.insert_fact(WorkingFact(
             fact_id="f_old", user_id="u1", subject="T",
+            session_id="test-session",
             statement="旧结论。", details="X=" + "b" * 8000,
         ))
         rec = WorkingFactReconciler(
@@ -782,6 +840,7 @@ class TestDetailsInheritance:
         facts, _ = await rec.reconcile(
             [WorkingFactDraft(subject="T", statement="新结论。", details="Y=1")],
             user_id="u1",
+            session_id="test-session",
         )
         stored = await coding_store.get_fact(facts[0].fact_id, user_id="u1")
         assert "…(truncated)" in stored.details
@@ -790,6 +849,7 @@ class TestDetailsInheritance:
     async def test_switch_off_keeps_details_untouched(self, coding_store, mock_embed):
         await coding_store.insert_fact(WorkingFact(
             fact_id="f_old", user_id="u1", subject="T",
+            session_id="test-session",
             statement="旧结论。", details="祖先取值 Z=9",
         ))
         rec = WorkingFactReconciler(
@@ -800,6 +860,7 @@ class TestDetailsInheritance:
         facts, ops = await rec.reconcile(
             [WorkingFactDraft(subject="T", statement="新结论。", details="Y=1")],
             user_id="u1",
+            session_id="test-session",
         )
         # 覆盖照样发生，只是不过继
         assert "SUPERSEDE" in [o.action for o in ops]
@@ -814,6 +875,7 @@ class TestDetailsInheritance:
         """
         await coding_store.insert_fact(WorkingFact(
             fact_id="f_old", user_id="u1", subject="T",
+            session_id="test-session",
             statement="旧结论。", details="无关噪声 " + "q" * 2000,
         ))
         rec = WorkingFactReconciler(
@@ -824,6 +886,7 @@ class TestDetailsInheritance:
         facts, _ = await rec.reconcile(
             [WorkingFactDraft(subject="T", statement="新结论。", details="Y=1")],
             user_id="u1",
+            session_id="test-session",
         )
         new_id = facts[0].fact_id
         emb = await coding_store.embed_service.embed("T 新结论。")

tests/coding/test_reconciler.py

@@ -35,7 +35,8 @@ class TestReconcilerActions:
     async def test_update(self, coding_store, mock_embed):
         # 种入 m1
         await coding_store.insert(CodingMemory(
-            experience_id="m1", user_id="u1", task="T1",
+            experience_id="m1", user_id="u1", agent_id="scope-day-1",
+            workspace_id="repo-a", session_id="conversation-1", task="T1",
             search_keys=["old"], solution="old",
         ))
         llm = MockLLMProvider(_llm_resp([
@@ -44,23 +45,146 @@ class TestReconcilerActions:
         ]))
         rec = CodingMemoryReconciler(coding_store, mock_embed, llm)
         d = CodingMemoryDraft(task="T1", search_keys=["new1"], solution="new",
-                              user_id="u1")
+                              user_id="u1", agent_id="scope-day-1",
+                              workspace_id="repo-a", session_id="conversation-1")
         ops = await rec.reconcile([d])
         assert len(ops) == 1 and ops[0].action == "UPDATE"
         got = await coding_store.get_by_id("m1", "u1")
         assert got.solution == "new"
         assert got.search_keys == ["new1"]
 
+    @pytest.mark.parametrize(
+        (
+            "old_agent",
+            "old_workspace",
+            "old_session",
+            "new_agent",
+            "new_workspace",
+            "new_session",
+        ),
+        [
+            (
+                "scope:same-day",
+                "repo-a",
+                "conversation-a",
+                "scope:same-day",
+                "repo-a",
+                "conversation-b",
+            ),
+            (
+                "scope:2026-10-18",
+                "repo-a",
+                "conversation-a",
+                "scope:2026-10-19",
+                "repo-a",
+                "conversation-a",
+            ),
+            (
+                "shared-scope",
+                "repo-a",
+                "conversation-a",
+                "shared-scope",
+                "repo-b",
+                "conversation-a",
+            ),
+        ],
+    )
+    async def test_update_target_must_match_frozen_authority_scope(
+        self,
+        coding_store,
+        mock_embed,
+        old_agent,
+        old_workspace,
+        old_session,
+        new_agent,
+        new_workspace,
+        new_session,
+    ):
+        """Near duplicates cannot cross conversation, date, or workspace authority."""
+        await coding_store.insert(CodingMemory(
+            experience_id="old-target", user_id="u1", agent_id=old_agent,
+            workspace_id=old_workspace, session_id=old_session,
+            task="fix flaky release", solution="old solution",
+        ))
+        llm = MockLLMProvider(_llm_resp([
+            {
+                "draft_idx": 0,
+                "action": "UPDATE",
+                "target_experience_id": "old-target",
+                "reason": "near duplicate",
+            },
+        ]))
+        rec = CodingMemoryReconciler(coding_store, mock_embed, llm)
+        draft = CodingMemoryDraft(
+            task="fix flaky release",
+            solution="new solution",
+            user_id="u1",
+            agent_id=new_agent,
+            workspace_id=new_workspace,
+            session_id=new_session,
+        )
+
+        ops = await rec.reconcile([draft])
+
+        old = await coding_store.get_by_id("old-target", "u1")
+        assert old is not None
+        assert old.solution == "old solution"
+        assert old.agent_id == old_agent
+        assert old.workspace_id == old_workspace
+        assert old.session_id == old_session
+        assert len(ops) == 1
+        assert ops[0].action == "ADD"
+        assert ops[0].target_experience_id != "old-target"
+
+    @pytest.mark.parametrize("legacy_agent", ["", "default_agent", "scope:legacy"])
+    async def test_legacy_target_without_conversation_falls_back_to_add(
+        self, coding_store, mock_embed, legacy_agent
+    ):
+        await coding_store.insert(CodingMemory(
+            experience_id="legacy-target",
+            user_id="u1",
+            agent_id=legacy_agent,
+            session_id=None,
+            workspace_id="repo-a",
+            task="upgrade dependency",
+            solution="legacy solution",
+        ))
+        llm = MockLLMProvider(_llm_resp([
+            {
+                "draft_idx": 0,
+                "action": "UPDATE",
+                "target_experience_id": "legacy-target",
+            },
+        ]))
+        rec = CodingMemoryReconciler(coding_store, mock_embed, llm)
+        draft = CodingMemoryDraft(
+            task="upgrade dependency",
+            solution="current solution",
+            user_id="u1",
+            agent_id=legacy_agent,
+            session_id="current-conversation",
+            workspace_id="repo-a",
+        )
+
+        ops = await rec.reconcile([draft])
+
+        legacy = await coding_store.get_by_id("legacy-target", "u1")
+        assert legacy is not None
+        assert legacy.solution == "legacy solution"
+        assert legacy.session_id is None
+        assert [op.action for op in ops] == ["ADD"]
+
     async def test_skip(self, coding_store, mock_embed):
         await coding_store.insert(CodingMemory(
             experience_id="m1", user_id="u1", task="X", solution="Y",
+            session_id="conversation-1",
         ))
         llm = MockLLMProvider(_llm_resp([
             {"draft_idx": 0, "action": "SKIP", "reason": "duplicate"},
         ]))
         rec = CodingMemoryReconciler(coding_store, mock_embed, llm)
         d = CodingMemoryDraft(task="X", solution="Y",
-                              user_id="u1")
+                              user_id="u1", session_id="conversation-1")
         ops = await rec.reconcile([d])
         assert len(ops) == 1 and ops[0].action == "SKIP"
         # store 不变
@@ -71,7 +195,7 @@ class TestReconcilerActions:
         # 旧记忆要被 DELETE
         await coding_store.insert(CodingMemory(
             experience_id="m_dep", user_id="u1", task="use old.sh",
-            solution="old script",
+            solution="old script", session_id="conversation-1",
         ))
         llm = MockLLMProvider(_llm_resp([
             {"draft_idx": 0, "action": "ADD"},
@@ -80,7 +204,7 @@ class TestReconcilerActions:
         ]))
         rec = CodingMemoryReconciler(coding_store, mock_embed, llm)
         d = CodingMemoryDraft(task="use new CLI", solution="new way",
-                              user_id="u1")
+                              user_id="u1", session_id="conversation-1")
         ops = await rec.reconcile([d])
         assert sorted(op.action for op in ops) == ["ADD", "DELETE"]
         # 旧的没了

tests/coding/test_store.py

@@ -67,7 +67,8 @@ class TestStoreCRUD:
 
     async def test_update_overrides(self, coding_store, mock_embed):
         m = CodingMemory(experience_id="e1", user_id="u1", task="T",
-                         search_keys=["old1", "old2"], solution="old")
+                         search_keys=["old1", "old2"], solution="old",
+                         session_id="conversation-1")
         await coding_store.insert(m)
 
         # 旧 aspect 应可命中
@@ -90,6 +91,59 @@ class TestStoreCRUD:
         assert got.solution == "new"
         assert got.search_keys == ["new1"]
 
+    async def test_update_rejects_authority_scope_change(self, coding_store):
+        await coding_store.insert(CodingMemory(
+            experience_id="e1", user_id="u1", agent_id="scope-a",
+            workspace_id="repo-a", session_id="conversation-1",
+            task="T", solution="old",
+        ))
+        replacement = CodingMemory(
+            experience_id="e1", user_id="u1", agent_id="scope-b",
+            workspace_id="repo-a", session_id="conversation-1",
+            task="T", solution="new",
+        )
+
+        with pytest.raises(ValueError, match="coding_memory_update_authority_mismatch"):
+            await coding_store.update(replacement)
+
+        stored = await coding_store.get_by_id("e1", user_id="u1")
+        assert stored is not None
+        assert stored.agent_id == "scope-a"
+        assert stored.solution == "old"
+
+    async def test_update_rejects_conversation_change_and_legacy_empty_session(
+        self, coding_store
+    ):
+        for experience_id, old_session in (
+            ("different-conversation", "conversation-1"),
+            ("legacy-empty-session", None),
+        ):
+            await coding_store.insert(CodingMemory(
+                experience_id=experience_id,
+                user_id="u1",
+                agent_id="scope-a",
+                workspace_id="repo-a",
+                session_id=old_session,
+                task="T",
+                solution="old",
+            ))
+            replacement = CodingMemory(
+                experience_id=experience_id,
+                user_id="u1",
+                agent_id="scope-a",
+                workspace_id="repo-a",
+                session_id="conversation-2",
+                task="T",
+                solution="new",
+            )
+
+            with pytest.raises(ValueError, match="coding_memory_update_authority_mismatch"):
+                await coding_store.update(replacement)
+
+            stored = await coding_store.get_by_id(experience_id, user_id="u1")
+            assert stored is not None
+            assert stored.solution == "old"
+
     async def test_delete(self, coding_store):
         m = CodingMemory(experience_id="e1", user_id="u1", task="T",
                          search_keys=["k"], solution="S")

tests/coding/test_writer_e2e.py

@@ -335,7 +335,7 @@ class TestDualLayerWrite:
         from hy_memory.coding.types import CodingMemory
         await coding_store.insert(CodingMemory(
             experience_id="e1", user_id="u1", task="老 task", solution="旧 solution",
-            source_fact_ids=["f_before"],
+            source_fact_ids=["f_before"], session_id="test-session",
         ))
         llm = MockLLMProvider([
             _extract_resp(
@@ -348,7 +348,9 @@ class TestDualLayerWrite:
         ])
         writer = self._writer(coding_store, mock_embed, llm)
 
-        resp = await writer.write(_coding_messages(), user_id="u1")
+        resp = await writer.write(
+            _coding_messages(), user_id="u1", session_id="test-session"
+        )
         exp = await coding_store.get_by_id("e1", user_id="u1")
         assert exp.solution == "新 solution"
         assert exp.source_fact_ids == ["f_before", resp["fact_ids"][0]]

tests/test_block_l1_raw_in_read_filter.py

@@ -12,8 +12,9 @@ L1_RAW 对外读路径不可见：闸门焊在向量库唯一查询入口。
 hybrid_search_native / list_by_user 四个查询方法的共同必经处，
 上层无法绕开，新增接口自动受保护。
 
-按 ID 直取走主键查询、不经过 filter，故在 `client.async_get` 处单独挡
+按 ID 直取走主键查询、不经过 filter，故在 `client.async_get` 处默认挡住。
-（GET /v1/memories/{id} 与 /v1/memories/batch 的共同必经处）。
+Python SDK 调用方可显式传 `include_raw=True` 读取；HTTP 接口不传该参数，
+继续保持原有隐藏行为。
 
 内部管线需读原始对话时显式 allow_raw_layer=True 放行：
 - `writer._get_recent_dialogue_history`：提取前拼多轮对话历史喂 LLM
@@ -124,28 +125,63 @@ class TestListNoLongerFiltersInProcess:
             "list 又出现进程内过滤 raw 的写法，应交由 vdb 侧 filter 完成"
 
 
-class TestGetByIdPathBlocksRaw:
+class TestGetByIdRawOptIn:
-    """按 ID 直取：走主键查询不经 filter，须在 async_get 处挡住。"""
+    """按 ID 直取默认隐藏 raw，只有 SDK 显式选择时才返回。"""
 
-    def test_async_get_drops_raw_node(self):
+    @staticmethod
+    def _client(node):
+        from hy_memory.client import HyMemoryClient
+
+        class Store:
+            async def get_by_id(self, memory_id):
+                return node if memory_id == node.node_id else None
+
+        async def no_op_log(**kwargs):
+            return None
+
+        client = HyMemoryClient.__new__(HyMemoryClient)
+        client._vector_store = Store()
+        client._graph_store = None
+        client._backend_identity = lambda: {}
+        client._maybe_store_op_log = no_op_log
+        return client
+
+    @staticmethod
+    def _raw_node():
+        return MemoryNode(
+            node_id="raw-1",
+            user_id="u1",
+            agent_id="a1",
+            session_id="s1",
+            content="原始对话",
+            layer=MemoryLayer.L1_RAW,
+            status=MemoryStatus.ACTIVE,
+            custom={"meta": [{"source": "test"}]},
+        )
+
+    def test_default_still_hides_raw(self):
+        from hy_memory.client import HyMemoryClient
+
+        client = self._client(self._raw_node())
+        raw_get = HyMemoryClient.async_get.__wrapped__
+        assert run(raw_get(client, "raw-1")) is None
+
+    def test_explicit_include_raw_returns_lite_memory(self):
+        from hy_memory.client import HyMemoryClient
+
+        client = self._client(self._raw_node())
+        raw_get = HyMemoryClient.async_get.__wrapped__
+        result = run(raw_get(client, "raw-1", include_raw=True))
+
+        assert result["memory_id"] == "raw-1"
+        assert result["content"] == "原始对话"
+        assert result["layer"] == MemoryLayer.L1_RAW.value
+        assert result["meta"] == [{"source": "test"}]
+
+    def test_public_get_exposes_keyword_only_opt_in(self):
         import inspect
         from hy_memory.client import HyMemoryClient
-        src = inspect.getsource(HyMemoryClient.async_get)
-        assert "MemoryLayer.L1_RAW" in src, \
-            "async_get 未挡 raw，batch 与按 ID 单取会漏出原始对话"
 
-    def test_block_happens_before_return(self):
+        param = inspect.signature(HyMemoryClient.get).parameters["include_raw"]
-        """挡的位置必须在两处 return 之前，否则等于没挡。"""
+        assert param.default is False
-        import inspect
+        assert param.kind is inspect.Parameter.KEYWORD_ONLY
-        from hy_memory.client import HyMemoryClient
-        lines = inspect.getsource(HyMemoryClient.async_get).splitlines()
-        block_at = next(
-            i for i, l in enumerate(lines)
-            if "MemoryLayer.L1_RAW" in l and "vdb_node" in l
-        )
-        first_return_at = next(
-            i for i, l in enumerate(lines)
-            if l.strip().startswith("return {")
-        )
-        assert block_at < first_return_at, \
-            f"挡 raw 在第 {block_at} 行、首个 return 在第 {first_return_at} 行，顺序错了"

tests/test_sqlite_connection_lifecycle.py

@@ -0,0 +1,136 @@
+import asyncio
+import threading
+
+import pytest
+
+from hy_memory.config import MemoryConfig
+from hy_memory.data.cache_sqlite import SqliteCache
+from hy_memory.data.history_store import HistoryStore
+
+
+class _BlockingCursor:
+    def fetchone(self):
+        return (1,)
+
+    def fetchall(self):
+        return []
+
+
+class _BlockingConnection:
+    """Hold execute() open so a concurrent close() can be observed."""
+
+    def __init__(self):
+        self.entered = threading.Event()
+        self.release = threading.Event()
+        self.closed = threading.Event()
+
+    def execute(self, *_args, **_kwargs):
+        self.entered.set()
+        if not self.release.wait(timeout=2):
+            raise TimeoutError("test did not release the blocking query")
+        if self.closed.is_set():
+            raise AssertionError("connection was closed while execute() was active")
+        return _BlockingCursor()
+
+    def close(self):
+        self.closed.set()
+
+
+def _config(tmp_path) -> MemoryConfig:
+    config = MemoryConfig()
+    config.history.db_path = str(tmp_path / "history.db")
+    config.cache.db_path = str(tmp_path / "cache.db")
+    return config
+
+
+async def _wait_for(event: threading.Event) -> None:
+    assert await asyncio.wait_for(asyncio.to_thread(event.wait, 1), timeout=2)
+
+
+@pytest.mark.asyncio
+async def test_history_close_waits_for_active_query(tmp_path):
+    store = HistoryStore(_config(tmp_path))
+    connection = _BlockingConnection()
+    store._conn = connection
+
+    query = asyncio.create_task(store.count())
+    await _wait_for(connection.entered)
+    closing = asyncio.create_task(store.close())
+    await asyncio.sleep(0.05)
+
+    assert not connection.closed.is_set()
+    connection.release.set()
+    assert await query == 1
+    await closing
+    assert connection.closed.is_set()
+
+
+@pytest.mark.asyncio
+async def test_cache_close_waits_for_active_query(tmp_path):
+    store = SqliteCache(_config(tmp_path))
+    connection = _BlockingConnection()
+    store._conn = connection
+
+    query = asyncio.create_task(store.load_metrics_range("0", "9"))
+    await _wait_for(connection.entered)
+    closing = asyncio.create_task(store.close())
+    await asyncio.sleep(0.05)
+
+    assert not connection.closed.is_set()
+    connection.release.set()
+    assert await query == []
+    await closing
+    assert connection.closed.is_set()
+
+
+@pytest.mark.asyncio
+async def test_history_concurrent_reads_and_writes_are_serialized(tmp_path):
+    store = HistoryStore(_config(tmp_path))
+    await store.initialize()
+    try:
+        inserted = await asyncio.gather(
+            *(
+                store.record_add(
+                    memory_id=f"memory-{index}",
+                    content="content",
+                    layer="l2_fact",
+                    isolation_key="user",
+                )
+                for index in range(200)
+            )
+        )
+        assert all(row_id > 0 for row_id in inserted)
+        assert len(set(inserted)) == 200
+
+        counts = await asyncio.gather(*(store.count() for _ in range(100)))
+        assert counts == [200] * 100
+    finally:
+        await store.close()
+
+
+@pytest.mark.asyncio
+async def test_cache_concurrent_pipeline_log_reads_and_writes_are_serialized(tmp_path):
+    store = SqliteCache(_config(tmp_path))
+    await store.initialize()
+    try:
+        written = await asyncio.gather(
+            *(
+                store.store_pipeline_log(
+                    request_id="request",
+                    user_id="user",
+                    agent_id="agent",
+                    step="EXTRACT",
+                    prompt=f"prompt-{index}",
+                    response="response",
+                )
+                for index in range(200)
+            )
+        )
+        assert written == [True] * 200
+
+        pages = await asyncio.gather(
+            *(store.get_pipeline_logs(request_id="request", limit=200) for _ in range(50))
+        )
+        assert all(len(page) == 200 for page in pages)
+    finally:
+        await store.close()