memorix 1.5.0 → 1.6.0

package/CHANGELOG.md

@@ -2,6 +2,43 @@
 
 All notable changes to this project will be documented in this file.
 
+## [Unreleased]
+
+## [1.6.0] - 2026-08-17
+
+### Added
+- **Bounded context receipts** -- `memorix context|resume --brief-json`,
+  `memorix codegraph context-pack --brief-json`, and MCP project-context
+  `format: "receipt"` return only the agent-ready brief, selection receipt,
+  CodeGraph quality, and optional Agent loadout. The detailed `--json` and
+  MCP `format: "json"` forms remain available for diagnostics.
+- **Task-scoped Agent loadouts** -- `--agent <target>` selects only compatible
+  project workflows and reports their required context and workflow tools.
+  This is guidance, not a replacement for the host agent's authorization.
+- **Explicit external CodeGraph lifecycle** -- `memorix codegraph init|sync`
+  manages an already installed local CodeGraph index. It never invokes
+  CodeGraph's agent installer or writes Agent configuration.
+
+### Changed
+- **Honest CodeGraph status** -- Status now distinguishes the persistent Lite
+  Code State index from the provider actually usable for the current task.
+  A stale external graph falls back to Lite instead of being presented as fresh.
+- **Bounded fallback guidance** -- Newly generated Agent rules teach
+  `--brief-json` for the one CLI fallback; `--json` is reserved for deliberate
+  detailed diagnostics.
+
+## [1.5.1] - 2026-08-16
+
+### Fixed
+- **Bounded formation cancellation** -- Formation deadlines now abort the
+  LLM and vector-search work they started, including response-body reads and
+  retry waits. A provider that ignores cancellation still cannot hold the
+  caller past its deadline.
+- **Embedding error classification** -- Only an explicit HTTP 400 batch-shape
+  error may trigger batch splitting. Authentication, billing, quota, and
+  upstream errors fail once instead of multiplying latency through recursive
+  retries.
+
 ## [1.5.0] - 2026-08-14
 
 ### Added

package/README.md

@@ -70,10 +70,10 @@ Memorix is more than a memory store. It also installs agent integrations, keeps
 
 | Capability | What it does | Entry point |
 | --- | --- | --- |
-| Memory Autopilot | A bounded task Workset with start files, current memory, source-backed knowledge, workflow starts, cautions, and verification | `memorix context "..."`, `memorix resume "..."`, `memorix_project_context` |
+| Memory Autopilot | A bounded task Workset with start files, current memory, source-backed knowledge, workflow starts, cautions, and verification. Compact JSON receipts keep fallback CLI use bounded too. | `memorix context "..." --brief-json`, `memorix resume "..." --brief-json`, `memorix_project_context` |
 | Observation Memory | Searchable facts, fixes, gotchas, session summaries, and implementation notes scoped to the current Git project | `memorix memory`, MCP memory tools |
 | Curated Long-term Memory | Deliberately reviewed episodic, semantic, and procedural memory with source evidence. Only an explicitly portable user item can cross local projects | `memorix memory long-term` |
-| Code State and Code Memory | Versioned local code snapshots, file/symbol links, and freshness checks. The built-in Lite index is always honest about its limits; an already-indexed local CodeGraph can add a bounded semantic outline | `memorix codegraph`, automatic context refresh |
+| Code State and Code Memory | Versioned local code snapshots, file/symbol links, and freshness checks. The built-in Lite index is always honest about its limits; an already-indexed local CodeGraph can add a bounded semantic outline and is only used when fresh. | `memorix codegraph status\|init\|sync`, automatic context refresh |
 | Git Memory | Commit-derived engineering facts that answer what changed, where, and why it matters | `memorix ingest commit`, git hook |
 | Reasoning Memory | Design rationale, alternatives, trade-offs, and risks that should survive beyond one chat | `memorix reasoning`, memory formation |
 | Knowledge Workspace | Review-gated source-backed claims, Markdown pages, and canonical project workflows; proposals never overwrite reviewed pages silently | `memorix knowledge`, `memorix knowledge workflow` |
@@ -437,7 +437,7 @@ Search is project-scoped by default. `scope="global"` searches across projects.
 
 Long-term memory is deliberately not an automatic dump of every note. A source observation, Claim, workflow, session, and code snapshot keep their existing roles. An agent may ask `memorix_store` to create an additional long-term record; an explicit request auto-qualifies and enters task briefs as a durable anchor, while hook-captured or Git-derived candidates stay pending until qualified. Approval stays an explicit operator review. Use `memorix memory long-term approve|archive|supersede` to record the evidence-backed lifecycle. Only a manually created or user-confirmed `user + portable` item may be considered in another local project; project code, Git facts, tests, workflows, sessions, and observations cannot be promoted into portable user memory.
 
-`memorix context "..."` is the default Memory Autopilot entry. It builds a compact task-lensed brief for agents: bugfix tasks lean toward tests and repros, release tasks lean toward package/changelog/build checks, onboarding tasks lean toward docs and entry points, and stale or unrelated memories stay in warning lanes instead of flooding the prompt. A normal new task does not receive an old-session dump. For an explicit continuation, `memorix resume "..."` adds only the latest useful session summary, up to three readable durable anchors, and at most one recent source-labelled host compact checkpoint. A durable anchor carries a `durable:<id>` reference, so an agent can expand the full reviewed record through `memorix_detail` only when needed. Keyword matches stay primary; when no reviewed durable item matches and an embedding provider is configured, Memorix makes one 1.8-second, no-retry semantic fallback for paraphrases or cross-language tasks. A slow or unavailable provider simply leaves the normal keyword-only Workset intact. A checkpoint is lifecycle evidence, not durable memory or a transcript backup. Agents should read the suggested files before trusting stored memory.
+`memorix context "..."` is the default Memory Autopilot entry. It builds a compact task-lensed brief for agents: bugfix tasks lean toward tests and repros, release tasks lean toward package/changelog/build checks, onboarding tasks lean toward docs and entry points, and stale or unrelated memories stay in warning lanes instead of flooding the prompt. Use `--brief-json` when a CLI fallback needs the same bounded brief plus a receipt of what was selected or omitted; `--json` remains the detailed diagnostic form. A normal new task does not receive an old-session dump. For an explicit continuation, `memorix resume "..."` adds only the latest useful session summary, up to three readable durable anchors, and at most one recent source-labelled host compact checkpoint. A durable anchor carries a `durable:<id>` reference, so an agent can expand the full reviewed record through `memorix_detail` only when needed. Keyword matches stay primary; when no reviewed durable item matches and an embedding provider is configured, Memorix makes one 1.8-second, no-retry semantic fallback for paraphrases or cross-language tasks. A slow or unavailable provider simply leaves the normal keyword-only Workset intact. A checkpoint is lifecycle evidence, not durable memory or a transcript backup. Agents should read the suggested files before trusting stored memory.
 
 <h2 id="runtime-modes"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-runtime.svg"><img src="assets/tags/section-runtime.svg" alt="Runtime Modes" height="32" /></picture></h2>
 

package/README.zh-CN.md

@@ -69,10 +69,10 @@ Memorix 不只是一个记忆库。它还负责安装 Agent 接入、保留有
 
 | 能力 | 作用 | 入口 |
 | --- | --- | --- |
-| Memory Autopilot | 给新 Agent session 一份有预算的任务 Workset，包含起步文件、当前记忆、来源知识、工作流首步、风险提示和验证建议 | `memorix context "..."`、`memorix resume "..."`、`memorix_project_context` |
+| Memory Autopilot | 给新 Agent session 一份有预算的任务 Workset，包含起步文件、当前记忆、来源知识、工作流首步、风险提示和验证建议。CLI 回退可用紧凑 JSON 回执，不再读取整套内部对象。 | `memorix context "..." --brief-json`、`memorix resume "..." --brief-json`、`memorix_project_context` |
 | Observation Memory | 当前 Git 项目内可检索的事实、修复、坑点、session 摘要和实现记录 | `memorix memory`、MCP memory tools |
 | 受管理的长期记忆 | 有来源证据、可审核的情景/语义/程序记忆；只有明确标为可携带的用户记忆才可在本机跨项目使用 | `memorix memory long-term` |
-| Code State 和 Code Memory | 可版本化的本地代码快照、文件 / symbol 关联和 freshness 检查。内置 Lite 会如实说明边界；已有本地索引的 CodeGraph 可额外给出有预算的语义关系 | `memorix codegraph`、自动 context refresh |
+| Code State 和 Code Memory | 可版本化的本地代码快照、文件 / symbol 关联和 freshness 检查。内置 Lite 会如实说明边界；已有本地索引的 CodeGraph 可额外给出有预算的语义关系，但只有新鲜时才会用于任务。 | `memorix codegraph status\|init\|sync`、自动 context refresh |
 | Git Memory | 从 commit 中提取工程事实，回答改了什么、在哪里改、为什么重要 | `memorix ingest commit`、git hook |
 | Reasoning Memory | 保存设计原因、备选方案、trade-off 和风险，不让决策只留在一次聊天里 | `memorix reasoning`、memory formation |
 | Knowledge Workspace | 有审核门槛的来源证据 Claim、Markdown 知识页和规范化项目工作流；提案不会悄悄覆盖已审阅页面 | `memorix knowledge`、`memorix knowledge workflow` |
@@ -420,7 +420,7 @@ memcode
 
 长期记忆不是把所有笔记自动堆进去。Observation、Claim、工作流、session 和代码快照仍各自承担原来的职责。Agent 可以在 `memorix_store` 时要求额外生成长期记忆**候选**，但候选绝不会自动进入任务上下文；用 `memorix memory long-term qualify|approve|archive|supersede` 留下带证据的生命周期记录。只有人工创建或用户确认的 `user + portable` 记忆，才可能在同一台机器的其它项目中按任务相关性被使用；项目代码、Git 事实、测试、工作流、session 和 Observation 都不能被提升成可携带的用户记忆。
 
-`memorix context "..."` 是默认的 Memory Autopilot 入口。它会按任务生成紧凑 brief：修 bug 时偏向测试和复现，发版时偏向 package/changelog/build 检查，接手项目时偏向文档和入口文件；过期或不相关的记忆只作为 warning，不会一股脑塞进 prompt。普通新任务不会自动得到旧会话的文本倾倒；明确要继续之前工作时，用 `memorix resume "..."`，只会补入最近一份有用的会话总结、最多三条当前可读的长期记忆锚点，以及最多一条带来源标识的近期宿主压缩检查点。每个长期锚点都有 `durable:<id>` 引用，Agent 只在确实需要完整已审核记录时才通过 `memorix_detail` 展开。关键词仍优先；没有命中而且用户已配置 embedding 时，Memorix 才会做一次 1.8 秒、不重试的语义回退，用于模型改写或跨语言任务。服务变慢或不可用时，正常的关键词 Workset 会原样返回。检查点只是生命周期证据，不是长期记忆，也不是聊天记录备份。Agent 应该先读 suggested files，再相信历史记忆。
+`memorix context "..."` 是默认的 Memory Autopilot 入口。它会按任务生成紧凑 brief：修 bug 时偏向测试和复现，发版时偏向 package/changelog/build 检查，接手项目时偏向文档和入口文件；过期或不相关的记忆只作为 warning，不会一股脑塞进 prompt。CLI 回退需要结构化结果时用 `--brief-json`，它只返回同一份有预算的 brief 和“选了什么、略了什么”的回执；`--json` 保留给详细诊断。普通新任务不会自动得到旧会话的文本倾倒；明确要继续之前工作时，用 `memorix resume "..."`，只会补入最近一份有用的会话总结、最多三条当前可读的长期记忆锚点，以及最多一条带来源标识的近期宿主压缩检查点。每个长期锚点都有 `durable:<id>` 引用，Agent 只在确实需要完整已审核记录时才通过 `memorix_detail` 展开。关键词仍优先；没有命中而且用户已配置 embedding 时，Memorix 才会做一次 1.8 秒、不重试的语义回退，用于模型改写或跨语言任务。服务变慢或不可用时，正常的关键词 Workset 会原样返回。检查点只是生命周期证据，不是长期记忆，也不是聊天记录备份。Agent 应该先读 suggested files，再相信历史记忆。
 
 <h2 id="运行模式"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-runtime.svg"><img src="assets/tags/section-runtime.svg" alt="运行模式" height="32" /></picture></h2>