# repomemory - Doramagic AI Context Pack

> 定位：安装前体验与判断资产。它帮助宿主 AI 有一个好的开始，但不代表已经安装、执行或验证目标项目。

## 充分原则

- **充分原则，不是压缩原则**：AI Context Pack 应该充分到让宿主 AI 在开工前理解项目价值、能力边界、使用入口、风险和证据来源；它可以分层组织，但不以最短摘要为目标。
- **压缩策略**：只压缩噪声和重复内容，不压缩会影响判断和开工质量的上下文。

## 给宿主 AI 的使用方式

你正在读取 Doramagic 为 repomemory 编译的 AI Context Pack。请把它当作开工前上下文：帮助用户理解适合谁、能做什么、如何开始、哪些必须安装后验证、风险在哪里。不要声称你已经安装、运行或执行了目标项目。

## Claim 消费规则

- **事实来源**：Repo Evidence + Claim/Evidence Graph；Human Wiki 只提供显著性、术语和叙事结构。
- **事实最低状态**：`supported`
- `supported`：可以作为项目事实使用，但回答中必须引用 claim_id 和证据路径。
- `weak`：只能作为低置信度线索，必须要求用户继续核实。
- `inferred`：只能用于风险提示或待确认问题，不能包装成项目事实。
- `unverified`：不得作为事实使用，应明确说证据不足。
- `contradicted`：必须展示冲突来源，不得替用户强行选择一个版本。

## 它最适合谁

- **正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**：README 或插件配置提到多个宿主 AI。 证据：`README.md` Claim：`clm_0004` supported 0.86
- **希望把专业流程带进宿主 AI 的用户**：仓库包含 Skill 文档。 证据：`skills/repomemory/SKILL.md`, `skills/session-end/SKILL.md`, `skills/session-start/SKILL.md` Claim：`clm_0005` supported 0.86

## 它能做什么

- **AI Skill / Agent 指令资产库**（可做安装前预览）：项目包含可被宿主 AI 读取的 Skill 或 Agent 指令文件，可用于把专业流程带入 Claude、Codex、Cursor 等宿主。 证据：`skills/repomemory/SKILL.md`, `skills/session-end/SKILL.md`, `skills/session-start/SKILL.md` Claim：`clm_0001` supported 0.86
- **多宿主安装与分发**（需要安装后验证）：项目包含插件或 marketplace 配置，说明它面向一个或多个 AI 宿主的安装和分发。 证据：`.claude-plugin/plugin.json` Claim：`clm_0002` supported 0.86
- **命令行启动或安装流程**（需要安装后验证）：项目文档中存在可执行命令，真实使用需要在本地或宿主环境中运行这些命令。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md` Claim：`clm_0003` supported 0.86

## 怎么开始

- `npx repomemory go` 证据：`README.md` Claim：`clm_0006` supported 0.86, `clm_0009` supported 0.86
- `npx repomemory setup cursor` 证据：`README.md` Claim：`clm_0007` supported 0.86
- `npx repomemory wizard` 证据：`README.md` Claim：`clm_0008` supported 0.86
- `npx repomemory go --yes --provider anthropic --embedding-provider gemini --max-files 80` 证据：`README.md` Claim：`clm_0009` supported 0.86
- `npx tsx src/index.ts --help` 证据：`AGENTS.md` Claim：`clm_0010` supported 0.86
- `npm install -g repomemory@latest && repomemory --version` 证据：`CLAUDE.md` Claim：`clm_0011` supported 0.86

## 继续前判断卡

- **当前建议**：需要管理员/安全审批
- **为什么**：继续前可能涉及密钥、账号、外部服务或敏感上下文，建议先经过管理员或安全审批。

### 30 秒判断

- **现在怎么做**：需要管理员/安全审批
- **最小安全下一步**：先跑 Prompt Preview；若涉及凭证或企业环境，先审批再试装
- **先别相信**：真实输出质量不能在安装前相信。
- **继续会触碰**：命令执行、宿主 AI 配置、本地环境或项目文件

### 现在可以相信

- **适合人群线索：正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0004` supported 0.86
- **适合人群线索：希望把专业流程带进宿主 AI 的用户**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`skills/repomemory/SKILL.md`, `skills/session-end/SKILL.md`, `skills/session-start/SKILL.md` Claim：`clm_0005` supported 0.86
- **能力存在：AI Skill / Agent 指令资产库**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`skills/repomemory/SKILL.md`, `skills/session-end/SKILL.md`, `skills/session-start/SKILL.md` Claim：`clm_0001` supported 0.86
- **能力存在：多宿主安装与分发**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`.claude-plugin/plugin.json` Claim：`clm_0002` supported 0.86
- **能力存在：命令行启动或安装流程**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md` Claim：`clm_0003` supported 0.86
- **存在 Quick Start / 安装命令线索**（supported）：可以相信项目文档出现过启动或安装入口；不要因此直接在主力环境运行。 证据：`README.md` Claim：`clm_0006` supported 0.86, `clm_0009` supported 0.86

### 现在还不能相信

- **真实输出质量不能在安装前相信。**（unverified）：Prompt Preview 只能展示引导方式，不能证明真实项目中的结果质量。
- **宿主 AI 版本兼容性不能在安装前相信。**（unverified）：Claude、Cursor、Codex、Gemini 等宿主加载规则和版本差异必须在真实环境验证。
- **不会污染现有宿主 AI 行为，不能直接相信。**（inferred）：Skill、plugin、AGENTS/CLAUDE/GEMINI 指令可能改变宿主 AI 的默认行为。 证据：`.claude-plugin/plugin.json`, `AGENTS.md`, `CLAUDE.md`, `skills/repomemory/SKILL.md` 等
- **可安全回滚不能默认相信。**（unverified）：除非项目明确提供卸载和恢复说明，否则必须先在隔离环境验证。
- **真实安装后是否与用户当前宿主 AI 版本兼容？**（unverified）：兼容性只能通过实际宿主环境验证。 证据：`.claude-plugin/plugin.json`
- **项目输出质量是否满足用户具体任务？**（unverified）：安装前预览只能展示流程和边界，不能替代真实评测。
- **安装命令是否需要网络、权限或全局写入？**（unverified）：这影响企业环境和个人环境的安装风险。 证据：`README.md`

### 继续会触碰什么

- **命令执行**：包管理器、网络下载、本地插件目录、项目配置或用户主目录。 原因：运行第一条命令就可能产生环境改动；必须先判断是否值得跑。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md`
- **宿主 AI 配置**：Claude/Codex/Cursor/Gemini/OpenCode 等宿主的 plugin、Skill 或规则加载配置。 原因：宿主配置会改变 AI 后续工作方式，可能和用户已有规则冲突。 证据：`.claude-plugin/plugin.json`, `AGENTS.md`, `CLAUDE.md`, `skills/repomemory/SKILL.md` 等
- **本地环境或项目文件**：安装结果、插件缓存、项目配置或本地依赖目录。 原因：安装前无法证明写入范围和回滚方式，需要隔离验证。 证据：`.claude-plugin/plugin.json`, `AGENTS.md`, `CLAUDE.md`, `README.md`
- **环境变量 / API Key**：项目入口文档明确出现 API key、token、secret 或账号凭证配置。 原因：如果真实安装需要凭证，应先使用测试凭证并经过权限/合规判断。 证据：`.context/facts/ai-providers.md`, `.context/facts/architecture.md`, `CLAUDE.md`, `README.md` 等
- **宿主 AI 上下文**：AI Context Pack、Prompt Preview、Skill 路由、风险规则和项目事实。 原因：导入上下文会影响宿主 AI 后续判断，必须避免把未验证项包装成事实。

### 最小安全下一步

- **先跑 Prompt Preview**：用安装前交互式试用判断工作方式是否匹配，不需要授权或改环境。（适用：任何项目都适用，尤其是输出质量未知时。）
- **只在隔离目录或测试账号试装**：避免安装命令污染主力宿主 AI、真实项目或用户主目录。（适用：存在命令执行、插件配置或本地写入线索时。）
- **先备份宿主 AI 配置**：Skill、plugin、规则文件可能改变 Claude/Cursor/Codex 的默认行为。（适用：存在插件 manifest、Skill 或宿主规则入口时。）
- **不要使用真实生产凭证**：环境变量/API key 一旦进入宿主或工具链，可能产生账号和合规风险。（适用：出现 API、TOKEN、KEY、SECRET 等环境线索时。）
- **安装后只验证一个最小任务**：先验证加载、兼容、输出质量和回滚，再决定是否深用。（适用：准备从试用进入真实工作流时。）

### 退出方式

- **保留安装前状态**：记录原始宿主配置和项目状态，后续才能判断是否可恢复。
- **准备移除宿主 plugin / Skill / 规则入口**：如果试装后行为异常，可以把宿主 AI 恢复到试装前状态。
- **记录安装命令和写入路径**：没有明确卸载说明时，至少要知道哪些目录或配置需要手动清理。
- **准备撤销测试 API key 或 token**：测试凭证泄露或误用时，可以快速止损。
- **如果没有回滚路径，不进入主力环境**：不可回滚是继续前阻断项，不应靠信任或运气继续。

## 哪些只能预览

- 解释项目适合谁和能做什么
- 基于项目文档演示典型对话流程
- 帮助用户判断是否值得安装或继续研究

## 哪些必须安装后验证

- 真实安装 Skill、插件或 CLI
- 执行脚本、修改本地文件或访问外部服务
- 验证真实输出质量、性能和兼容性

## 边界与风险判断卡

- **把安装前预览误认为真实运行**：用户可能高估项目已经完成的配置、权限和兼容性验证。 处理方式：明确区分 prompt_preview_can_do 与 runtime_required。 Claim：`clm_0012` inferred 0.45
- **宿主 AI 插件或 Skill 规则冲突**：新规则可能改变用户现有宿主 AI 的工作方式。 处理方式：安装前先检查插件 manifest 和 Skill 文件，必要时隔离测试。 证据：`.claude-plugin/plugin.json` Claim：`clm_0013` supported 0.86
- **命令执行会修改本地环境**：安装命令可能写入用户主目录、宿主插件目录或项目配置。 处理方式：先在隔离环境或测试账号中运行。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md` Claim：`clm_0014` supported 0.86
- **待确认**：真实安装后是否与用户当前宿主 AI 版本兼容？。原因：兼容性只能通过实际宿主环境验证。
- **待确认**：项目输出质量是否满足用户具体任务？。原因：安装前预览只能展示流程和边界，不能替代真实评测。
- **待确认**：安装命令是否需要网络、权限或全局写入？。原因：这影响企业环境和个人环境的安装风险。

## 开工前工作上下文

### 加载顺序

- 先读取 how_to_use.host_ai_instruction，建立安装前判断资产的边界。
- 读取 claim_graph_summary，确认事实来自 Claim/Evidence Graph，而不是 Human Wiki 叙事。
- 再读取 intended_users、capabilities 和 quick_start_candidates，判断用户是否匹配。
- 需要执行具体任务时，优先查 role_skill_index，再查 evidence_index。
- 遇到真实安装、文件修改、网络访问、性能或兼容性问题时，转入 risk_card 和 boundaries.runtime_required。

### 任务路由

- **AI Skill / Agent 指令资产库**：先基于 role_skill_index / evidence_index 帮用户挑选可用角色、Skill 或工作流。 边界：可做安装前 Prompt 体验。 证据：`skills/repomemory/SKILL.md`, `skills/session-end/SKILL.md`, `skills/session-start/SKILL.md` Claim：`clm_0001` supported 0.86
- **多宿主安装与分发**：先说明这是安装后验证能力，再给出安装前检查清单。 边界：必须真实安装或运行后验证。 证据：`.claude-plugin/plugin.json` Claim：`clm_0002` supported 0.86
- **命令行启动或安装流程**：先说明这是安装后验证能力，再给出安装前检查清单。 边界：必须真实安装或运行后验证。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md` Claim：`clm_0003` supported 0.86

### 上下文规模

- 文件总数：120
- 重要文件覆盖：40/120
- 证据索引条目：80
- 角色 / Skill 条目：3

### 证据不足时的处理

- **missing_evidence**：说明证据不足，要求用户提供目标文件、README 段落或安装后验证记录；不要补全事实。
- **out_of_scope_request**：说明该任务超出当前 AI Context Pack 证据范围，并建议用户先查看 Human Manual 或真实安装后验证。
- **runtime_request**：给出安装前检查清单和命令来源，但不要替用户执行命令或声称已执行。
- **source_conflict**：同时展示冲突来源，标记为待核实，不要强行选择一个版本。

## Prompt Recipes

### 适配判断

- 目标：判断这个项目是否适合用户当前任务。
- 预期输出：适配结论、关键理由、证据引用、安装前可预览内容、必须安装后验证内容、下一步建议。

```text
请基于 repomemory 的 AI Context Pack，先问我 3 个必要问题，然后判断它是否适合我的任务。回答必须包含：适合谁、能做什么、不能做什么、是否值得安装、证据来自哪里。所有项目事实必须引用 evidence_refs、source_paths 或 claim_id。
```

### 安装前体验

- 目标：让用户在安装前感受核心工作流，同时避免把预览包装成真实能力或营销承诺。
- 预期输出：一段带边界标签的体验剧本、安装后验证清单和谨慎建议；不含真实运行承诺或强营销表述。

```text
请把 repomemory 当作安装前体验资产，而不是已安装工具或真实运行环境。

请严格输出四段：
1. 先问我 3 个必要问题。
2. 给出一段“体验剧本”：用 [安装前可预览]、[必须安装后验证]、[证据不足] 三种标签展示它可能如何引导工作流。
3. 给出安装后验证清单：列出哪些能力只有真实安装、真实宿主加载、真实项目运行后才能确认。
4. 给出谨慎建议：只能说“值得继续研究/试装”“先补充信息后再判断”或“不建议继续”，不得替项目背书。

硬性边界：
- 不要声称已经安装、运行、执行测试、修改文件或产生真实结果。
- 不要写“自动适配”“确保通过”“完美适配”“强烈建议安装”等承诺性表达。
- 如果描述安装后的工作方式，必须使用“如果安装成功且宿主正确加载 Skill，它可能会……”这种条件句。
- 体验剧本只能写成“示例台词/假设流程”：使用“可能会询问/可能会建议/可能会展示”，不要写“已写入、已生成、已通过、正在运行、正在生成”。
- Prompt Preview 不负责给安装命令；如用户准备试装，只能提示先阅读 Quick Start 和 Risk Card，并在隔离环境验证。
- 所有项目事实必须来自 supported claim、evidence_refs 或 source_paths；inferred/unverified 只能作风险或待确认项。

```

### 角色 / Skill 选择

- 目标：从项目里的角色或 Skill 中挑选最匹配的资产。
- 预期输出：候选角色或 Skill 列表，每项包含适用场景、证据路径、风险边界和是否需要安装后验证。

```text
请读取 role_skill_index，根据我的目标任务推荐 3-5 个最相关的角色或 Skill。每个推荐都要说明适用场景、可能输出、风险边界和 evidence_refs。
```

### 风险预检

- 目标：安装或引入前识别环境、权限、规则冲突和质量风险。
- 预期输出：环境、权限、依赖、许可、宿主冲突、质量风险和未知项的检查清单。

```text
请基于 risk_card、boundaries 和 quick_start_candidates，给我一份安装前风险预检清单。不要替我执行命令，只说明我应该检查什么、为什么检查、失败会有什么影响。
```

### 宿主 AI 开工指令

- 目标：把项目上下文转成一次对话开始前的宿主 AI 指令。
- 预期输出：一段边界明确、证据引用明确、适合复制给宿主 AI 的开工前指令。

```text
请基于 repomemory 的 AI Context Pack，生成一段我可以粘贴给宿主 AI 的开工前指令。这段指令必须遵守 not_runtime=true，不能声称项目已经安装、运行或产生真实结果。
```


## 角色 / Skill 索引

- 共索引 3 个角色 / Skill / 项目文档条目。

- **repomemory**（skill）：Search and manage your repository's persistent knowledge base. Use this to find architecture docs, past decisions, known regressions, coding preferences, and session notes before starting work. 激活提示：当用户任务与“repomemory”描述的流程高度相关时，先用它做安装前体验，再决定是否安装。 证据：`skills/repomemory/SKILL.md`
- **session-end**（skill）：Record what you accomplished and discovered during this session. Routes conclusions to the right categories. 激活提示：当用户任务与“session-end”描述的流程高度相关时，先用它做安装前体验，再决定是否安装。 证据：`skills/session-end/SKILL.md`
- **session-start**（skill）：Orient yourself at the start of a coding session. Loads project context, recent sessions, and relevant knowledge automatically. 激活提示：当用户任务与“session-start”描述的流程高度相关时，先用它做安装前体验，再决定是否安装。 证据：`skills/session-start/SKILL.md`

## 证据索引

- 共索引 80 条证据。

- **repomemory — Instructions for AI Coding Agents**（documentation）：repomemory — Instructions for AI Coding Agents 证据：`AGENTS.md`
- **repomemory — Agent Instructions**（documentation）：This repo uses repomemory https://github.com/DanielGuru/repomemory for persistent AI memory. 证据：`CLAUDE.md`
- **repomemory**（documentation）：AI agents lose context every session. repomemory fixes that — one command creates a persistent, searchable knowledge base that any AI tool can read, search, and write to. 证据：`README.md`
- **Package**（package_manifest）：{ "name": "repomemory", "version": "1.11.0", "description": "Your codebase never forgets. Persistent, structured memory for AI coding agents.", "type": "module", "bin": { "repomemory": "dist/index.js" }, "exports": { ".": { "types": "./dist/index.d.ts", "default": "./dist/index.js" }, "./mcp": { "types": "./dist/mcp/server.d.ts", "default": "./dist/mcp/server.js" } }, "scripts": { "sync:versions": "node scripts/sync-server-version.js", "check:release": "node scripts/check-release-consistency.js", "build": "npm run sync:versions && node scripts/build.js", "typecheck": "tsc --noEmit", "format": "prettier --write 'src/ / .ts' 'tests/ / .ts' 'scripts/ / .js'", "format:check": "prettier --check… 证据：`package.json`
- **Contributing to repomemory**（documentation）：Thanks for your interest in contributing to repomemory! This guide will help you get set up and understand how the codebase works. 证据：`.github/CONTRIBUTING.md`
- **repomemory — Repository Memory**（skill_instruction）：You have access to the repomemory MCP server which provides persistent knowledge for this codebase. 证据：`skills/repomemory/SKILL.md`
- **Session End**（skill_instruction）：1. Summarize what was accomplished in this session 2. Route knowledge to the RIGHT categories — do NOT dump everything into sessions/: - New architectural facts - context write category="facts", ... - Decisions made - context write category="decisions", ... - Bugs/regressions found - context write category="regressions", ... - Coding style preferences - context write category="preferences", ... - The session overview itself - context write category="sessions", ... 3. Include: files changed, decisions made, issues encountered, things learned 4. Be specific: include file paths, function names, and commands 证据：`skills/session-end/SKILL.md`
- **Session Start**（skill_instruction）：1. Call context auto orient to get full project orientation in one call 2. Display the results: project overview, recent sessions, developer preferences, recently updated entries 3. If the context is empty, suggest running npx repomemory analyze 4. If the user mentioned a specific task, also run context search with relevant keywords 证据：`skills/session-start/SKILL.md`
- **Plugin**（structured_config）：{ "name": "repomemory", "version": "1.3.0", "description": "Persistent, structured memory for AI coding agents. Your codebase never forgets. Hybrid keyword + semantic search, auto-session capture, intelligent category routing.", "author": "DanielGuru", "homepage": "https://github.com/DanielGuru/repomemory", "license": "MIT" } 证据：`.claude-plugin/plugin.json`
- **License**（source_file）：Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files the "Software" , to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: 证据：`LICENSE`
- **repomemory — Agent Quick Reference**（documentation）：What This Is A CLI tool + MCP server npm package: repomemory that gives AI coding agents persistent, structured memory for repositories. Agents search, write, and delete knowledge that persists across sessions. 证据：`.context/index.md`
- **What This PR Does**（documentation）：A brief description of the changes in this pull request. 证据：`.github/PULL_REQUEST_TEMPLATE.md`
- **Changelog**（documentation）：New Features - Git intelligence — New repomemory risk CLI command and context risk MCP tool. Analyzes git history for hotspot files high churn , hidden coupling co-change detection via Jaccard similarity , ownership concentration, and bus factor - Risk-aware agent workflow — context auto orient now surfaces top risk hotspots at session start; start-task prompt instructs agents to call context risk before modifying files; CLAUDE MD BLOCK includes risk assessment in "Before modifying code" instructions 证据：`CHANGELOG.md`
- **Contributor Covenant Code of Conduct**（documentation）：Contributor Covenant Code of Conduct 证据：`CODE_OF_CONDUCT.md`
- **repomemory Launch Copy**（documentation）：repomemory Launch Copy Ready to copy-paste. Fill in DEMO GIF HERE once you have the recording. 证据：`LAUNCH-COPY.md`
- **Title**（documentation）：Title feat: harden onboarding/reliability + doctor diagnostics + CI/release consistency 证据：`PR-CODEX-HARDENING.md`
- **repomemory — Product + Technical Review Codex**（documentation）：repomemory — Product + Technical Review Codex 证据：`REPO-CODEX-REVIEW.md`
- **Security Policy**（documentation）：If you discover a security vulnerability in repomemory, please report it responsibly. 证据：`SECURITY.md`
- **Changelog — 2026-02**（documentation）：- 6031efe v1.3.0 — 53 bug fixes, security hardening, community files DanielGuru, 1 second ago - fc541e5 fix: add missing scope type to context write args DanielGuru, 31 minutes ago - 0438430 v1.2.0 — Global developer context, 5 bug fixes, dogfooded .context/ DanielGuru, 33 minutes ago - c78d958 v1.1.1: Update changelog, sync package-lock, commit pending docs DanielGuru, 2 hours ago - a7f93e0 v1.1.1 — Fix 25 audit findings: security, bugs, and hardening DanielGuru, 2 hours ago - 15969b3 v1.1.0 — Hybrid search, auto-session capture, intelligent routing, preferences DanielGuru, 2 hours ago - b14c211 Revert "Add v1.1 handover document for next development phase" DanielGuru, 3 hours ago - 1fa83a… 证据：`.context/changelog/2026-02.md`
- **Decision: Analysis Prompt V2 — Quality Over Quantity**（documentation）：Decision: Analysis Prompt V2 — Quality Over Quantity 证据：`.context/decisions/analysis-prompt-v2-design.md`
- **Decision: Custom Build Script Instead of Plain tsc**（documentation）：Decision: Custom Build Script Instead of Plain tsc 证据：`.context/decisions/custom-build-script.md`
- **Decision: TypeScript ESM + NodeNext Module Resolution**（documentation）：Decision: TypeScript ESM + NodeNext Module Resolution 证据：`.context/decisions/esm-nodenext.md`
- **Decision: Use Null Byte Delimiter for Git Log Parsing**（documentation）：Decision: Use Null Byte Delimiter for Git Log Parsing 证据：`.context/decisions/git-null-byte-delimiter.md`
- **Decision: Make go Command Interactive**（documentation）：Decision: Make go Command Interactive 证据：`.context/decisions/go-command-interactive.md`
- **Decision: Hybrid Keyword + Vector Search**（documentation）：Decision: Hybrid Keyword + Vector Search 证据：`.context/decisions/hybrid-search.md`
- **Decision: Extract JSON Repair to Separate Module**（documentation）：Decision: Extract JSON Repair to Separate Module 证据：`.context/decisions/json-repair-extraction.md`
- **Rejected Feature Ideas — 2026-02-18**（documentation）：Rejected Feature Ideas — 2026-02-18 证据：`.context/decisions/rejected-feature-ideas-2026-02-18.md`
- **Decision: Dirty Flag for Search Index Saves**（documentation）：Decision: Dirty Flag for Search Index Saves 证据：`.context/decisions/search-dirty-flag.md`
- **Decision: Session Capture Threshold — Drop Duration Requirement**（documentation）：Decision: Session Capture Threshold — Drop Duration Requirement 证据：`.context/decisions/session-capture-threshold.md`
- **Decision: Six Fixed Knowledge Categories**（documentation）：Decision: Six Fixed Knowledge Categories 证据：`.context/decisions/six-categories.md`
- **Decision: sql.js Wasm SQLite for Search**（documentation）：Decision: sql.js Wasm SQLite for Search 证据：`.context/decisions/sql-js-wasm.md`
- **v1.2: Global Context Layer**（documentation）：Decision Added a developer-layer global context at ~/.repomemory/global/ that persists across all repos. Preferences follow the developer; architecture/decisions/sessions stay repo-scoped. 证据：`.context/decisions/v12-global-context-layer.md`
- **AI Provider System**（documentation）：Supported Providers Provider Env Var Notes ---------- --------- ------- anthropic ANTHROPIC API KEY Uses streaming .stream not .create openai OPENAI API KEY Standard gemini GEMINI API KEY Uses systemInstruction field grok GROK API KEY OpenAI-compatible 证据：`.context/facts/ai-providers.md`
- **Architecture**（documentation）：Overview repomemory is a single-package CLI + MCP server. No microservices. Deployed as an npm package. 证据：`.context/facts/architecture.md`
- **Build & Deployment**（documentation）：Build Script: scripts/build.js - Custom build script not tsc directly - Compiles TypeScript to dist/ - Adds shebang to dist/index.js — critical for CLI usage - If dist/index.js lacks !/usr/bin/env node , check this script 证据：`.context/facts/build-deploy.md`
- **Claude Code Post-Commit Hook**（documentation）：What It Does After repomemory setup claude , a PostToolUse hook is installed at .claude/settings.json that fires after Bash commands containing git commit . It outputs a reminder to the agent to record context using context write . 证据：`.context/facts/claude-code-post-commit-hook.md`
- **CLI Search Command**（documentation）：Added in v1.4.0. Searches repo + global context from the terminal. 证据：`.context/facts/cli-search-command.md`
- **Doctor Command**（documentation）：repomemory doctor runs diagnostics and outputs a health report. 证据：`.context/facts/doctor-command.md`
- **Fact: Repo Scanner Dotdir Allowlist**（documentation）：Fact: Repo Scanner Dotdir Allowlist 证据：`.context/facts/dotdir-allowlist.md`
- **MCP Server Details**（documentation）：Session Tracking - Tracks all tool calls during a session - Auto-writes session summary to sessions/ on graceful shutdown SIGTERM/SIGINT - Works with ALL MCP clients — no hooks required 证据：`.context/facts/mcp-server.md`
- **Non-Interactive Setup**（documentation）：Both go and wizard support deterministic, non-interactive execution for CI and scripting. 证据：`.context/facts/non-interactive-setup.md`
- **Search System**（documentation）：Overview Hybrid search combining FTS5 keyword search sql.js Wasm SQLite with optional vector/semantic search. 证据：`.context/facts/search-system.md`
- **TypeScript Style Preferences**（documentation）：Strict Mode tsconfig.json has "strict": true — all strict TypeScript checks enabled. 证据：`.context/preferences/typescript-style.md`
- **Regression: Anthropic SDK Requires Streaming**（documentation）：Regression: Anthropic SDK Requires Streaming 证据：`.context/regressions/anthropic-streaming-required.md`
- **Regression: context write Called listEntries on Every Write**（documentation）：Regression: context write Called listEntries on Every Write 证据：`.context/regressions/context-write-listentries-overhead.md`
- **Regression: ESM require in Dashboard**（documentation）：Regression: ESM require in Dashboard 证据：`.context/regressions/dashboard-esm-require.md`
- **Regression: Dry-Run Mode Validated API Key Unnecessarily**（documentation）：Regression: Dry-Run Mode Validated API Key Unnecessarily 证据：`.context/regressions/dry-run-api-key-validation.md`
- **Regression: FTS5 Not Available in All sql.js Builds**（documentation）：Regression: FTS5 Not Available in All sql.js Builds 证据：`.context/regressions/fts5-fallback-missing.md`
- **Regression: Global Search Index Had No Lazy-Init Fallback**（documentation）：Regression: Global Search Index Had No Lazy-Init Fallback 证据：`.context/regressions/global-search-index-no-lazy-init.md`
- **Regression: init Command Auto-Modified CLAUDE.md**（documentation）：Regression: init Command Auto-Modified CLAUDE.md 证据：`.context/regressions/init-auto-modifying-claude-md.md`
- **Regression: Path Traversal in readEntry and listEntries**（documentation）：Regression: Path Traversal in readEntry and listEntries 证据：`.context/regressions/path-traversal-readentry.md`
- **Regression: searchLike Completely Non-Functional**（documentation）：Regression: searchLike Completely Non-Functional 证据：`.context/regressions/searchlike-broken-sql.md`
- **Regression: Session Auto-Capture Never Fires in Production**（documentation）：Regression: Session Auto-Capture Never Fires in Production 证据：`.context/regressions/session-autocapture-stdin-close.md`
- **Regression: setup claude Wrote to Wrong Config File**（documentation）：Regression: setup claude Wrote to Wrong Config File 证据：`.context/regressions/setup-claude-wrong-config-file.md`
- **Comprehensive Audit of src/commands/ — 2026-02-18**（documentation）：Comprehensive Audit of src/commands/ — 2026-02-18 证据：`.context/sessions/2026-02-18-commands-audit.md`
- **Comprehensive Audit Session — 2026-02-18**（documentation）：Comprehensive Audit Session — 2026-02-18 证据：`.context/sessions/2026-02-18-comprehensive-audit.md`
- **v1.2 Implementation Session — 2026-02-18**（documentation）：v1.2 Implementation Session — 2026-02-18 证据：`.context/sessions/2026-02-18-v12-implementation.md`
- **Session: Cursor MCP Integration**（documentation）：What Was Done Made repomemory Cursor-ready by adding full MCP server configuration, mirroring what setup claude already does. 证据：`.context/sessions/2026-02-19-cursor-mcp-integration.md`
- **v1.7.0 Release Session — 2026-02-20**（documentation）：v1.7.0 Release Session — 2026-02-20 证据：`.context/sessions/2026-02-20-v170-release.md`
- **v1.10.0 Release Session — 2026-02-22**（documentation）：v1.10.0 Release Session — 2026-02-22 证据：`.context/sessions/2026-02-22-v1100-release.md`
- 其余 20 条证据见 `AI_CONTEXT_PACK.json` 或 `EVIDENCE_INDEX.json`。

## 宿主 AI 必须遵守的规则

- **把本资产当作开工前上下文，而不是运行环境。**：AI Context Pack 只包含证据化项目理解，不包含目标项目的可执行状态。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md`
- **回答用户时区分可预览内容与必须安装后才能验证的内容。**：安装前体验的消费者价值来自降低误装和误判，而不是伪装成真实运行。 证据：`AGENTS.md`, `CLAUDE.md`, `README.md`

## 用户开工前应该回答的问题

- 你准备在哪个宿主 AI 或本地环境中使用它？
- 你只是想先体验工作流，还是准备真实安装？
- 你最在意的是安装成本、输出质量、还是和现有规则的冲突？

## 验收标准

- 所有能力声明都能回指到 evidence_refs 中的文件路径。
- AI_CONTEXT_PACK.md 没有把预览包装成真实运行。
- 用户能在 3 分钟内看懂适合谁、能做什么、如何开始和风险边界。

---

## Doramagic Context Augmentation

The following material strengthens the Repomix/AI Context Pack body. Human Manual is only a reading skeleton; pitfall logs become hard operating constraints for the host AI.

## Human Manual Skeleton

Usage rule: this is only a reading path and salience signal, not factual authority. Concrete facts must still come from repo evidence / Claim Graph.

Hard rules for the host AI:
- Do not treat page titles, order, summaries, or importance as project facts.
- When explaining the Human Manual skeleton, state that it is only a reading path / salience signal.
- Capability, installation, compatibility, runtime status, and risk judgments must cite repo evidence, source paths, or Claim Graph.

- **Repository Overview**：importance `high`
  - source_paths: README.md, package.json, src/index.ts, src/commands/analyze.ts, src/commands/dashboard.ts
- **Entrypoints and Runtime Surface**：importance `high`
  - source_paths: README.md, package.json, src/index.ts, src/commands/analyze.ts, src/commands/dashboard.ts
- **Architecture Evidence Map**：importance `high`
  - source_paths: README.md, package.json, src/index.ts, src/commands/analyze.ts, src/commands/dashboard.ts
- **Operations and Verification Boundaries**：importance `high`
  - source_paths: README.md, package.json, src/index.ts, src/commands/analyze.ts, src/commands/dashboard.ts

## Repo Inspection Evidence

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `b5b9d25a1ba13198eaaa2bd672e6181deeb5dee9`
- inspected_files: `package.json`, `README.md`, `src/index.ts`, `src/mcp/server.ts`, `src/lib/search.ts`, `src/lib/git.ts`, `src/lib/ai-provider.ts`, `src/lib/config.ts`, `src/lib/git-intelligence.ts`, `src/lib/embeddings.ts`, `src/lib/context-store.ts`, `src/lib/repo-scanner.ts`, `src/lib/json-repair.ts`, `src/types/sql.js.d.ts`, `src/commands/wizard.ts`, `src/commands/global.ts`, `src/commands/hook.ts`, `src/commands/doctor.ts`, `src/commands/analyze.ts`, `src/commands/search.ts`

Hard rules for the host AI:
- Without repo_clone_verified=true, do not claim the source code has been read.
- Without repo_inspection_verified=true, do not turn README/docs/package observations into facts.
- Without quick_start_verified=true, do not claim the Quick Start has been successfully run.

## Doramagic Pitfall Constraints

These rules come from Doramagic discovery, validation, or compilation pitfalls. The host AI must treat them as operating constraints, not general background notes.

### Constraint 1: 能力判断依赖假设

- Trigger: README/documentation is current enough for a first validation pass.
- Host AI rule: 将假设转成下游验证清单。
- Why it matters: 假设不成立时，用户拿不到承诺的能力。
- Evidence: capability.assumptions | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | README/documentation is current enough for a first validation pass.
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.

### Constraint 2: 维护活跃度未知

- Trigger: 未记录 last_activity_observed。
- Host AI rule: 补 GitHub 最近 commit、release、issue/PR 响应信号。
- Why it matters: 新项目、停更项目和活跃项目会被混在一起，推荐信任度下降。
- Evidence: evidence.maintainer_signals | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | last_activity_observed missing
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.

### Constraint 3: 下游验证发现风险项

- Trigger: no_demo
- Host AI rule: 进入安全/权限治理复核队列。
- Why it matters: 下游已经要求复核，不能在页面中弱化。
- Evidence: downstream_validation.risk_items | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | no_demo; severity=medium
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.

### Constraint 4: 存在评分风险

- Trigger: no_demo
- Host AI rule: 把风险写入边界卡，并确认是否需要人工复核。
- Why it matters: 风险会影响是否适合普通用户安装。
- Evidence: risks.scoring_risks | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | no_demo; severity=medium
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.

### Constraint 5: issue/PR 响应质量未知

- Trigger: issue_or_pr_quality=unknown。
- Host AI rule: 抽样最近 issue/PR，判断是否长期无人处理。
- Why it matters: 用户无法判断遇到问题后是否有人维护。
- Evidence: evidence.maintainer_signals | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | issue_or_pr_quality=unknown
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.

### Constraint 6: 发布节奏不明确

- Trigger: release_recency=unknown。
- Host AI rule: 确认最近 release/tag 和 README 安装命令是否一致。
- Why it matters: 安装命令和文档可能落后于代码，用户踩坑概率升高。
- Evidence: evidence.maintainer_signals | mcp_registry:io.github.DanielGuru/repomemory:1.3.0 | https://registry.modelcontextprotocol.io/v0.1/servers/io.github.DanielGuru%2Frepomemory/versions/1.3.0 | release_recency=unknown
- Hard boundary: do not present this pitfall as solved, verified, or safe to ignore unless later validation evidence explicitly closes it.