# hunch - 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 hunch 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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude/skills/fable-mode/SKILL.md`, `plugin/skills/capture/SKILL.md` 等 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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude/skills/fable-mode/SKILL.md`, `plugin/skills/capture/SKILL.md` 等 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/marketplace.json`, `plugin/.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：`README.md` Claim：`clm_0003` supported 0.86

## 怎么开始

- `npm i -g @davesheffer/hunch` 证据：`README.md` Claim：`clm_0006` supported 0.86

## 继续前判断卡

- **当前建议**：先做权限沙盒试用
- **为什么**：项目存在安装命令、宿主配置或本地写入线索，不建议直接进入主力环境，应先在隔离环境试装。

### 30 秒判断

- **现在怎么做**：先做权限沙盒试用
- **最小安全下一步**：先跑 Prompt Preview；若仍要安装，只在隔离环境试装
- **先别相信**：工具权限边界不能在安装前相信。
- **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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude/skills/fable-mode/SKILL.md`, `plugin/skills/capture/SKILL.md` 等 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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude/skills/fable-mode/SKILL.md`, `plugin/skills/capture/SKILL.md` 等 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/marketplace.json`, `plugin/.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：`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

### 现在还不能相信

- **工具权限边界不能在安装前相信。**（unverified）：MCP/tool 类项目通常会触碰文件、网络、浏览器或外部 API，必须真实检查权限和日志。
- **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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude-plugin/marketplace.json`, `.claude/skills/fable-mode/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/marketplace.json`, `plugin/.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：`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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude-plugin/marketplace.json`, `.claude/skills/fable-mode/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/marketplace.json`, `README.md`, `plugin/.claude-plugin/plugin.json`
- **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。）
- **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。
- **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_0007` 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/marketplace.json`, `plugin/.claude-plugin/plugin.json` Claim：`clm_0008` 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：`README.md` Claim：`clm_0009` 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：`.agents/skills/fable-mode/SKILL.md`, `.agents/skills/source-command-hunch-fragile/SKILL.md`, `.claude/skills/fable-mode/SKILL.md`, `plugin/skills/capture/SKILL.md` 等 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/marketplace.json`, `plugin/.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：`README.md` Claim：`clm_0003` supported 0.86

### 上下文规模

- 文件总数：483
- 重要文件覆盖：40/483
- 证据索引条目：79
- 角色 / Skill 条目：8

### 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 hunch 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 hunch 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 hunch 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 8 个角色 / Skill / 项目文档条目。

- **fable-mode**（skill）：MANDATORY before diagnosing any bug, debugging any failure, reviewing code, or making a non-trivial change — invoke FIRST, before reading code or proposing a cause; do not skip because the task "looks simple". Five-gate rigor protocol scope, evidence, attack, verify, report with per-task playbooks and tripwires. Also fires on "fable mode", "be rigorous", "do this properly", "think hard", and any task where a wrong a… 激活提示：当用户任务与“fable-mode”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：`.agents/skills/fable-mode/SKILL.md`
- **source-command-hunch-fragile**（skill）：Report the most fragile parts of this codebase, with evidence 激活提示：当用户任务与“source-command-hunch-fragile”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：`.agents/skills/source-command-hunch-fragile/SKILL.md`
- **fable-mode**（skill）：MANDATORY before diagnosing any bug, debugging any failure, reviewing code, or making a non-trivial change — invoke FIRST, before reading code or proposing a cause; do not skip because the task "looks simple". Five-gate rigor protocol scope, evidence, attack, verify, report with per-task playbooks and tripwires. Also fires on "fable mode", "be rigorous", "do this properly", "think hard", and any task where a wrong a… 激活提示：当用户任务与“fable-mode”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：`.claude/skills/fable-mode/SKILL.md`
- **Skill**（skill）：Capture an engineering decision into Hunch's graph via a grilling interview topic, rationale, rejected alternatives 激活提示：当用户任务与“Skill”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：`plugin/skills/capture/SKILL.md`
- **Skill**（skill）：Fix a bug grounded in Hunch past root causes, constraints, blast radius 激活提示：当用户任务与“Skill”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：`plugin/skills/fix/SKILL.md`
- **Skill**（skill）：Report the most fragile parts of this codebase, with evidence from Hunch 激活提示：当用户任务与“Skill”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：`plugin/skills/fragile/SKILL.md`
- **Skill**（skill）：Reconcile docs/code with Hunch's decision graph doc≠graph drift , never rewriting prose silently 激活提示：当用户任务与“Skill”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：`plugin/skills/heal/SKILL.md`
- **Skill**（skill）：Explain why a file or symbol is the way it is, from Hunch's decision graph 激活提示：当用户任务与“Skill”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：`plugin/skills/why/SKILL.md`

## 证据索引

- 共索引 79 条证据。

- **Hunch**（documentation）：Your AI can write code. Hunch makes it remember the consequences. 证据：`README.md`
- **bench — does the skill / the graph actually help?**（documentation）：bench — does the skill / the graph actually help? 证据：`bench/README.md`
- **Hunch for VS Code**（documentation）：Your repo's engineering memory Hunch https://github.com/davesheffer/hunch , in the editor. One read surface, one write surface, nothing else: 证据：`vscode-extension/README.md`
- **Package**（package_manifest）：{ "name": "@davesheffer/hunch", "version": "1.7.1", "license": "Apache-2.0", "author": "Dave Sheffer ", "description": "Architectural Conformance for AI-generated code: a git-native graph that deterministically blocks AI changes which break your architecture — the semantic invariants layering, must-reach, dependency direction pattern-SAST can't express — grounded in the decisions and bugs behind each rule, across any MCP assistant Claude Code, Cursor, Copilot, Windsurf, Antigravity, Codex .", "homepage": "https://hunch-pi.vercel.app", "repository": { "type": "git", "url": "git+https://github.com/davesheffer/hunch.git" }, "bugs": { "url": "https://github.com/davesheffer/hunch/issues" }, "typ… 证据：`package.json`
- **Package**（package_manifest）：{ "name": "hunch-vscode", "displayName": "Hunch", "description": "Your repo's engineering memory, in the editor: see WHY code is the way it is decisions, invariants, bug history , and capture what you decide — one command each.", "version": "0.14.0", "publisher": "davesheffer", "license": "MIT", "author": "Dave Sheffer ", "icon": "media/icon.png", "homepage": "https://hunch-pi.vercel.app", "repository": { "type": "git", "url": "git+https://github.com/davesheffer/hunch.git", "directory": "vscode-extension" }, "bugs": { "url": "https://github.com/davesheffer/hunch/issues" }, "galleryBanner": { "color": " 16150f", "theme": "dark" }, "keywords": "claude-code", "mcp", "engineering-memory", "know… 证据：`vscode-extension/package.json`
- **AGENTS.md**（documentation）：This repo has Hunch — a curated graph of why the code is the way it is decisions, bug history, invariants . It currently holds 101 decisions, 1 bugs, 8 constraints, 16 components, 3 policies . 证据：`AGENTS.md`
- **CLAUDE.md**（documentation）：This file provides guidance to Claude Code claude.ai/code when working with code in this repository. 证据：`CLAUDE.md`
- **Contributing to Hunch**（documentation）：Thanks for helping build the Engineering Memory OS. 证据：`CONTRIBUTING.md`
- **Fable Mode**（skill_instruction）：Each gate exists to kill one specific failure mode. The gates shape the work; they are invisible in the output — only Gate 5 is what the user sees. Never narrate "now entering Gate 2." A failed gate sends you backward, never forward. 证据：`.agents/skills/fable-mode/SKILL.md`
- **source-command-hunch-fragile**（skill_instruction）：Use this skill when the user asks to run the migrated source command hunch-fragile . 证据：`.agents/skills/source-command-hunch-fragile/SKILL.md`
- **Fable Mode**（skill_instruction）：Each gate exists to kill one specific failure mode. The gates shape the work; they are invisible in the output — only Gate 5 is what the user sees. Never narrate "now entering Gate 2." A failed gate sends you backward, never forward. 证据：`.claude/skills/fable-mode/SKILL.md`
- **Skill**（skill_instruction）：Capture the decision for $ARGUMENTS into Hunch's graph. 证据：`plugin/skills/capture/SKILL.md`
- **Skill**（skill_instruction）：Follow the Hunch-grounded workflow — do NOT skip the memory lookups: 1. hunch bug lineage "$ARGUMENTS" — has this class of bug happened before? what was the root cause and the fix? 2. Identify the suspect symbol/file, then hunch get dependents to learn the blast radius. 3. hunch check constraints — list invariants you must preserve. 4. Propose a fix that honors past root causes AND constraints. Apply it and run the tests. 5. If the fix encodes a non-trivial choice, hunch record decision ... so the next session is grounded in it. 证据：`plugin/skills/fix/SKILL.md`
- **Skill**（skill_instruction）：Ask Hunch for the fragility ranking run hunch fragile or query Hunch , then produce a fragility report with evidence : the specific files/functions, the bug history behind them, their churn and fan-in, and any missing guards. Avoid generic advice — every claim must cite a Hunch record or metric. 证据：`plugin/skills/fragile/SKILL.md`
- **Skill**（skill_instruction）：Reconcile decision-grounding drift for $ARGUMENTS or the whole repo . 证据：`plugin/skills/heal/SKILL.md`
- **Skill**（skill_instruction）：Use the hunch why MCP tool on $ARGUMENTS a file path or symbol name . 证据：`plugin/skills/why/SKILL.md`
- **Marketplace**（structured_config）：{ "name": "hunch", "owner": { "name": "Dave Sheffer", "email": "dave.sheffer1@gmail.com" }, "description": "Hunch — Engineering Memory OS: a git-native reasoning graph of decisions, bugs, and invariants, enforced at reasoning time.", "plugins": { "name": "hunch", "source": "./plugin", "description": "The hunch MCP tools why, constraints, blast radius, merge verdict, capture plus /hunch:capture, /hunch:heal, /hunch:why, /hunch:fix, /hunch:fragile. Requires Node 22.13+; runs npx @davesheffer/hunch mcp per project.", "author": { "name": "Dave Sheffer", "email": "dave.sheffer1@gmail.com" }, "homepage": "https://hunch-pi.vercel.app", "repository": "https://github.com/davesheffer/hunch", "license… 证据：`.claude-plugin/marketplace.json`
- **Plugin**（structured_config）：{ "name": "hunch", "displayName": "Hunch — Engineering Memory OS", "description": "A git-native reasoning graph of decisions, bugs, and invariants for your codebase — queried and enforced at reasoning time. Adds the hunch MCP tools plus /hunch:capture, /hunch:heal, /hunch:why, /hunch:fix, /hunch:fragile.", "author": { "name": "Dave Sheffer", "email": "dave.sheffer1@gmail.com", "url": "https://github.com/davesheffer" }, "homepage": "https://hunch-pi.vercel.app", "repository": "https://github.com/davesheffer/hunch", "license": "Apache-2.0", "keywords": "engineering-memory", "architectural-conformance", "decisions", "invariants", "mcp", "knowledge-graph" , "mcpServers": "./.mcp.json", "skills"… 证据：`plugin/.claude-plugin/plugin.json`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`LICENSE`
- **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：`vscode-extension/LICENSE`
- **Hunch Cookbook**（documentation）：Practical recipes for running Hunch day to day. Each recipe is copy-paste-able and states what you should observe when it works. 证据：`docs/cookbook.md`
- **Capture**（documentation）：Capture the decision for $ARGUMENTS into Hunch's graph. 证据：`.claude/commands/capture.md`
- **Heal**（documentation）：Reconcile decision-grounding drift for $ARGUMENTS or the whole repo . 证据：`.claude/commands/heal.md`
- **Hunch Fix**（documentation）：Follow the Hunch-grounded workflow DESIGN §5 — do NOT skip the memory lookups: 1. hunch bug lineage "$ARGUMENTS" — has this class of bug happened before? what was the root cause and the fix? 2. Identify the suspect symbol/file, then hunch get dependents to learn the blast radius. 3. hunch check constraints — list invariants you must preserve. 4. Propose a fix that honors past root causes AND constraints. Apply it and run the tests. 5. If the fix encodes a non-trivial choice, hunch record decision ... so the next session is grounded in it. 证据：`.claude/commands/hunch-fix.md`
- **Hunch Fragile**（documentation）：Ask Hunch for the fragility ranking run hunch fragile or query Hunch , then produce a fragility report with evidence : the specific files/functions, the bug history behind them, their churn and fan-in, and any missing guards. Avoid generic advice — every claim must cite a Hunch record or metric. 证据：`.claude/commands/hunch-fragile.md`
- **Hunch Why**（documentation）：Use the hunch why MCP tool on $ARGUMENTS a file path or symbol name . 证据：`.claude/commands/hunch-why.md`
- **Index**（source_file）：import { existsSync, readFileSync, writeFileSync, mkdirSync, realpathSync } from "node:fs"; import { execFileSync, spawnSync } from "node:child process"; import { join, relative, dirname, basename, resolve, isAbsolute } from "node:path"; import { fileURLToPath } from "node:url"; import { Command } from "commander"; import { hunchPaths, hunchPathsForDir, findRoot, toPosixTarget } from "../core/paths.js"; import { writeFileAtomic } from "../core/io.js"; import { looksLikeCorrection, CORRECTION NUDGE } from "../core/correction.js"; import { HUNCH VERSION } from "../core/version.js"; import { HunchStore } from "../store/hunchStore.js"; import { JsonStore } from "../store/jsonStore.js"; import {… 证据：`src/cli/index.ts`
- **Invocation**（source_file）：import { fileURLToPath } from "node:url"; import type { Invocation } from "../integrations/scaffold.js"; ⋮---- export interface ResolvedInvocation { shell: string; mcp: Invocation; } ⋮---- export function resolveInvocation : ResolvedInvocation ⋮---- const q = s: string 证据：`src/cli/invocation.ts`
- **Adapters**（source_file）：import { execFileSync } from "node:child process"; import { readFileSync, statSync } from "node:fs"; import { basename, resolve } from "node:path"; import { shortHash } from "../core/ids.js"; import { parseDocAnchors } from "../core/docanchors.js"; import { toPosixTarget } from "../core/paths.js"; import type { Bug, Constraint } from "../core/types.js"; import { commitMeta, revExists, revParse } from "../extractors/git.js"; import type { HunchStore } from "../store/hunchStore.js"; import { durationCutoff } from "./bootstrap.js"; import { canonicalHash } from "./canonical.js"; import type { PolicyRepository } from "./repository.js"; import { EvidenceEventSchema, EvidenceImportSchema, type Da… 证据：`src/constitution/adapters.ts`
- **Lifecycle**（source_file）：import { assertCompositionBinding, exceptionScopeIsNarrower, oppositeExceptionAssertions, policyProofHash } from "./composition.js"; import { assessHistoryDispositions } from "./disposition.js"; import { type HistoryDisposition, type PolicyProof, type PolicySpec, type ProofClass } from "./schema.js"; import { evaluatorForPolicy, mutationEngineForPolicy } from "./policyRuntime.js"; import { executableBehaviorAttestationError } from "./behaviorAttestationBinding.js"; import type { G2BehaviorAttestation } from "./g2BehaviorAttestation.js"; ⋮---- function isHumanActor actor: string : boolean ⋮---- export function blockingEvidenceError proof: PolicyProof, dispositions: HistoryDisposition = : str… 证据：`src/constitution/lifecycle.ts`
- **Dupdetect**（source_file）：import type { Decision } from "./types.js"; import { toPosixTarget } from "./paths.js"; ⋮---- export function dupTerms text: string : Set ⋮---- function jaccard a: Set , b: Set : number ⋮---- function fileSet files: readonly string undefined : Set ⋮---- export interface Coverage { id: string; title: string; hoursAgo: number; fileOverlapPct: number; } ⋮---- export function commitCoveredBy codeFiles: readonly string , subject: string, existing: readonly Decision , nowMs: number, : Coverage null ⋮---- export interface DupMatch { of: Decision; score: number; } ⋮---- export function isAcceptedDuplicateAnchor d: Decision : boolean ⋮---- export function draftDuplicateOf draft: Decision, existing:… 证据：`src/core/dupdetect.ts`
- **Strictgate**（source_file）：export function isHumanConfirmed source: string undefined : boolean ⋮---- export interface StrictConstraint { severity?: string; provenance?: { confidence?: number; source?: string }; } ⋮---- export function isStrictBlocker c: StrictConstraint, stale: boolean : boolean ⋮---- export type VetoTier = "dep" "symbol" "pattern" "semantic"; ⋮---- export function isVetoBlocker d: { status?: string; superseded by?: string null }, tw: { provenance?: { source?: string } }, tier: VetoTier, stale: boolean, : boolean 证据：`src/core/strictgate.ts`
- **Version**（source_file）：import { readFileSync } from "node:fs"; import { fileURLToPath } from "node:url"; import { dirname, join } from "node:path"; 证据：`src/core/version.ts`
- **Hooks**（source_file）：import { readFileSync, writeFileSync, existsSync, chmodSync, mkdirSync } from "node:fs"; import { join, isAbsolute } from "node:path"; import { hooksDir } from "../extractors/git.js"; ⋮---- function block invocation: string, opts: ⋮---- export interface HookInstall { path: string; action: "created" "appended" "updated" "unchanged"; } ⋮---- export function installPostCommitHook root: string, invocation: string, opts: ⋮---- function escapeRe s: string : string ⋮---- export function installPreCommitHook root: string, invocation: string, strict = false : HookInstall 证据：`src/integrations/hooks.ts`
- **Provider**（source_file）：import { spawn } from "node:child process"; import { existsSync, mkdirSync, readFileSync } from "node:fs"; import { tmpdir } from "node:os"; import { dirname, join } from "node:path"; import { writeFileAtomic } from "../core/io.js"; import { summarizeDiff, type DiffAnalysis } from "../extractors/diff.js"; import type { Decision } from "../core/types.js"; ⋮---- export function pexecIn cmd: string, args: string , opts: { input?: string; env?: NodeJS.ProcessEnv; cwd?: string; timeout?: number; maxBuffer?: number; } = {}, : Promise void = ⋮---- export interface DecisionDraft { title: string; context: string; decision: string; consequences: string ; alternatives rejected: string ; confidence: nu… 证据：`src/synthesis/provider.ts`
- **Cli**（source_file）：export interface CliResult { ok: boolean; stdout: string; stderr: string; code: number null; } ⋮---- export function cliCommand : string ⋮---- export function winQuote a: string : string ⋮---- export function runHunch root: string, args: string , timeoutMs = 120 000 : Promise ⋮---- const settle = resolve: r: CliResult = void = err: cp.ExecException cp.ExecFileException null, stdout: string, stderr: string = ⋮---- export interface HunchProc { result: Promise ; kill : void; } ⋮---- export function spawnHunchProc root: string, args: string , onLine: line: string = void, timeoutMs = 900 000, : HunchProc ⋮---- let kill: = void = = ⋮---- kill = = ⋮---- if buf.trim onLine buf ; // flush a trailing… 证据：`vscode-extension/src/cli.ts`
- **Extension**（source_file）：import { loadHunch, why, constraintsInScope, nearConstraints, symbolSignals, bugsForSymbol, type Hunch, } from "./hunchData.js"; import { HunchHoverProvider } from "./providers.js"; import { runSearch } from "./search.js"; import { showJourney, resolveWikiGraph } from "./journey.js"; import { cliCommand, runHunchWithProgress } from "./cli.js"; import { registerLmTools } from "./lmTools.js"; import { HunchMcp } from "./mcpClient.js"; ⋮---- function workspaceRoot : string undefined ⋮---- function relPath file: string : string ⋮---- class HunchCache ⋮---- constructor private root: string undefined reload : Hunch null get : Hunch null ⋮---- function esc s: string : string ⋮---- function showBri… 证据：`vscode-extension/src/extension.ts`
- **Journey**（source_file）：import { reviewQueue, type Hunch } from "./hunchData.js"; import { runHunch, cliCommand } from "./cli.js"; ⋮---- interface Stats { stock?: { coverage?: { pct?: number } }; return?: { lifetime?: Record }; compounding?: { rules recorded?: number; catches lifetime?: number; payback ratio?: number }; } ⋮---- interface GraphDoc { rel: string; title: string; status: "grounded" "stale" "unverified"; adopted: string null; components: string ; } interface GraphData { kind: "public" "private"; nodes: Array ; links: Array ; docs: GraphDoc ; pendingReview: number; } ⋮---- export function resolveWikiGraph root: string, overlay?: ⋮---- const wikiDir = hunchDir: string : string = ⋮---- function readGraphD… 证据：`vscode-extension/src/journey.ts`
- **Lmtools**（source_file）：import { why, nearConstraints, constraintsInScope, searchAll, fragileSymbols, type Hunch, } from "./hunchData.js"; ⋮---- type GetHunch = = Hunch null; ⋮---- function normTarget target: string, root: string undefined : string ⋮---- function text s: string : vscode.LanguageModelToolResult ⋮---- / Resolve a target that may be a symbol name rather than a file path. / function fileForTarget hunch: Hunch, target: string : ⋮---- function whyText hunch: Hunch, rawTarget: string : string ⋮---- const section = h: string, items: string : void = ⋮---- class WhyTool implements vscode.LanguageModelTool string undefined prepareInvocation o: vscode.LanguageModelToolInvocationPrepareOptions string undefined… 证据：`vscode-extension/src/lmTools.ts`
- **Mcpclient**（source_file）：import { cliCommand, winQuote } from "./cli.js"; ⋮---- interface Pending { resolve: v: unknown = void; reject: e: Error = void; } ⋮---- export class HunchMcp ⋮---- constructor private root: string ⋮---- private start : Promise ⋮---- const die = why: string = ⋮---- private request method: string, params: unknown : Promise ⋮---- private notify method: string, params: unknown : void ⋮---- async call tool: string, args: unknown : Promise ⋮---- dispose : void 证据：`vscode-extension/src/mcpClient.ts`
- **Security Policy**（documentation）：Email dave.sheffer1@gmail.com or open a private security advisory https://github.com/davesheffer/hunch/security/advisories/new . Please do not file public issues for undisclosed vulnerabilities. Expect an acknowledgement within a few days. 证据：`SECURITY.md`
- **Architectural Conformance — prevent + catch benchmark**（documentation）：Architectural Conformance — prevent + catch benchmark 证据：`bench/architectural-conformance.md`
- **We let the AI audit its own memory tool. It was using 4 of 19 tools.**（documentation）：We let the AI audit its own memory tool. It was using 4 of 19 tools. 证据：`blog/agent-audits-its-own-memory.md`
- **AI ignores your architecture rules — even at the frontier. We measured it.**（documentation）：AI ignores your architecture rules — even at the frontier. We measured it. 证据：`blog/ai-ignores-your-architecture.md`
- **Fable Mode — Backend playbook**（documentation）：APIs, services, data, contracts, migrations. The gates, sharpened for state you can corrupt and callers you can't see. 证据：`.agents/skills/fable-mode/references/backend.md`
- **Fable Mode — Frontend playbook**（documentation）：UI, components, styling, browser behavior. The gates, sharpened for pixels. 证据：`.agents/skills/fable-mode/references/frontend.md`
- **Fable Mode — Infra / DevOps playbook**（documentation）：Fable Mode — Infra / DevOps playbook 证据：`.agents/skills/fable-mode/references/infra.md`
- **Fable Mode — QA / test playbook**（documentation）：Writing tests, fixing tests, building test plans. The prime directive: a test's value is its ability to FAIL. A test you never saw fail is a hope, not a test. 证据：`.agents/skills/fable-mode/references/qa.md`
- **Fable Mode — Verdict playbook**（documentation）：Design decisions, red teams, "which option is right." Not for execution work — this playbook manufactures a CALL, and its byproducts are the record: real rejected alternatives, evidenced attacks, flip conditions. 证据：`.agents/skills/fable-mode/references/verdict.md`
- **Soul**（documentation）：PIPELINE ACTIVE — operating loop enforced by hooks, not optional : 证据：`.claude/pipeline/soul.md`
- **Fable Mode — Backend playbook**（documentation）：APIs, services, data, contracts, migrations. The gates, sharpened for state you can corrupt and callers you can't see. 证据：`.claude/skills/fable-mode/references/backend.md`
- **Fable Mode — Frontend playbook**（documentation）：UI, components, styling, browser behavior. The gates, sharpened for pixels. 证据：`.claude/skills/fable-mode/references/frontend.md`
- **Fable Mode — Infra / DevOps playbook**（documentation）：Fable Mode — Infra / DevOps playbook 证据：`.claude/skills/fable-mode/references/infra.md`
- **Fable Mode — QA / test playbook**（documentation）：Writing tests, fixing tests, building test plans. The prime directive: a test's value is its ability to FAIL. A test you never saw fail is a hope, not a test. 证据：`.claude/skills/fable-mode/references/qa.md`
- **Fable Mode — Verdict playbook**（documentation）：Design decisions, red teams, "which option is right." Not for execution work — this playbook manufactures a CALL, and its byproducts are the record: real rejected alternatives, evidenced attacks, flip conditions. 证据：`.claude/skills/fable-mode/references/verdict.md`
- **🧠 Hunch Engineering Memory**（documentation）：This repo has Hunch — a curated graph of why the code is the way it is decisions, bug history, invariants . It currently holds 101 decisions, 1 bugs, 8 constraints, 16 components, 3 policies . 证据：`.windsurf/rules/hunch.md`
- **Capture what you decide**（documentation）：When you make a non-trivial choice, record it so the next person or agent inherits it. 证据：`vscode-extension/media/walkthrough/record.md`
- **Why is this the way it is?**（documentation）：- Hover a symbol to see its bug history and fragility. - The status bar shows how many invariants guard the active file — click it for the full brief. 证据：`vscode-extension/media/walkthrough/why.md`
- **Hooks**（structured_config）：{ "hooks": { "PreToolUse": { "matcher": "Edit Write MultiEdit", "hooks": { "type": "command", "command": "\"/Users/nofarkulishevski/.hermes/node/bin/node\" \"/Users/nofarkulishevski/.local/lib/node modules/@davesheffer/hunch/dist/cli/index.js\" hook" } } , "PostToolUse": { "matcher": "Edit Write MultiEdit Bash PowerShell Skill", "hooks": { "type": "command", "command": "\"/Users/nofarkulishevski/.hermes/node/bin/node\" \"/Users/nofarkulishevski/.local/lib/node modules/@davesheffer/hunch/dist/cli/index.js\" hook" } } , "SessionStart": { "hooks": { "type": "command", "command": "\"/Users/nofarkulishevski/.hermes/node/bin/node\" \"/Users/nofarkulishevski/.local/lib/node modules/@davesheffer/hu… 证据：`.codex/hooks.json`
- **Manifest**（structured_config）：{ "schema version": 2 } 证据：`.hunch/manifest.json`
- 其余 19 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：`README.md`, `bench/README.md`, `vscode-extension/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：`README.md`, `bench/README.md`, `vscode-extension/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

下面内容用于强化 Repomix/AI Context Pack 主体。Human Manual 只提供阅读骨架；踩坑日志会被转成宿主 AI 必须遵守的工作约束。

## Human Manual 骨架

使用规则：这里只是项目阅读路线和显著性信号，不是事实权威。具体事实仍必须回到 repo evidence / Claim Graph。

宿主 AI 硬性规则：
- 不得把页标题、章节顺序、摘要或 importance 当作项目事实证据。
- 解释 Human Manual 骨架时，必须明确说它只是阅读路线/显著性信号。
- 能力、安装、兼容性、运行状态和风险判断必须引用 repo evidence、source path 或 Claim Graph。

- **项目概览**：importance `high`
  - source_paths: README.md, bench/README.md, package.json, vscode-extension/README.md, vscode-extension/package.json
- **Src 模块**：importance `high`
  - source_paths: vscode-extension/src/cli.ts, vscode-extension/src/extension.ts, vscode-extension/src/hunchData.ts, vscode-extension/src/journey.ts, vscode-extension/src/lmTools.ts
- **Cli 模块**：importance `high`
  - source_paths: src/cli/index.ts, src/cli/invocation.ts, src/cli/preflight.ts
- **Commands 模块**：importance `high`
  - source_paths: .claude/commands/capture.md, .claude/commands/heal.md, .claude/commands/hunch-fix.md, .claude/commands/hunch-fragile.md, .claude/commands/hunch-why.md
- **Index.ts 模块**：importance `medium`
  - source_paths: src/cli/index.ts
- **Invocation.ts 模块**：importance `medium`
  - source_paths: src/cli/invocation.ts
- **Preflight.ts 模块**：importance `medium`
  - source_paths: src/cli/preflight.ts
- **Adapters.ts 模块**：importance `medium`
  - source_paths: src/constitution/adapters.ts

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `62842da2bf949d50099eb16b16322428cfa6f983`
- inspected_files: `README.md`, `package.json`, `docs/cookbook.md`, `docs/evidence/constitution-phase-3-clean-install.json`, `src/cli/index.ts`, `src/cli/invocation.ts`, `src/cli/preflight.ts`, `src/constitution/adapters.ts`, `src/constitution/behaviorAttestationBinding.ts`, `src/constitution/behaviorEvaluator.ts`, `src/constitution/behaviorProof.ts`, `src/constitution/behaviorWorkspace.ts`, `src/constitution/bootstrap.ts`, `src/constitution/canonical.ts`, `src/constitution/card.ts`, `src/constitution/compiler.ts`, `src/constitution/composition.ts`, `src/constitution/corpus.ts`, `src/constitution/delta.ts`, `src/constitution/disposition.ts`

宿主 AI 硬性规则：
- 没有 repo_clone_verified=true 时，不得声称已经读过源码。
- 没有 repo_inspection_verified=true 时，不得把 README/docs/package 文件判断写成事实。
- 没有 quick_start_verified=true 时，不得声称 Quick Start 已跑通。

## Doramagic Pitfall Constraints / 踩坑约束

这些规则来自 Doramagic 发现、验证或编译过程中的项目专属坑点。宿主 AI 必须把它们当作工作约束，而不是普通说明文字。

### Constraint 1: 可能修改宿主 AI 配置

- Trigger: 项目面向 Claude/Cursor/Codex/Gemini/OpenCode 等宿主，或安装命令涉及用户配置目录。
- Host AI rule: 列出会写入的配置文件、目录和卸载/回滚步骤。
- Why it matters: 安装可能改变本机 AI 工具行为，用户需要知道写入位置和回滚方法。
- Evidence: capability.host_targets | https://github.com/davesheffer/hunch | host_targets=mcp_host, claude_code, claude, cursor
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 2: 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 | https://github.com/davesheffer/hunch | README/documentation is current enough for a first validation pass.
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 3: 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 | https://github.com/davesheffer/hunch | last_activity_observed missing
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

- Trigger: no_demo
- Evidence: downstream_validation.risk_items | https://github.com/davesheffer/hunch | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 5: 存在评分风险

- Trigger: no_demo
- Why it matters: 风险会影响是否适合普通用户安装。
- Evidence: risks.scoring_risks | https://github.com/davesheffer/hunch | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 6: 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 | https://github.com/davesheffer/hunch | issue_or_pr_quality=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 7: 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 | https://github.com/davesheffer/hunch | release_recency=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。