# caveman - 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 caveman 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

- **AI 研究者或研究型 Agent 构建者**：README 明确围绕研究、实验或论文工作流展开。 证据：`README.md` Claim：`clm_0004` supported 0.86
- **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_0005` 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：`plugins/caveman/skills/cavecrew/SKILL.md`, `plugins/caveman/skills/caveman/SKILL.md`, `plugins/caveman/skills/caveman-compress/SKILL.md`, `plugins/caveman/skills/caveman-stats/SKILL.md` 等 Claim：`clm_0006` 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：`plugins/caveman/skills/cavecrew/SKILL.md`, `plugins/caveman/skills/caveman/SKILL.md`, `plugins/caveman/skills/caveman-compress/SKILL.md`, `plugins/caveman/skills/caveman-stats/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`, `.claude-plugin/plugin.json`, `plugins/caveman/.codex-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：`INSTALL.md`, `README.md` Claim：`clm_0003` supported 0.86

## 怎么开始

- `curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bash` 证据：`INSTALL.md` Claim：`clm_0007` supported 0.86, `clm_0008` supported 0.86, `clm_0014` supported 0.86
- `curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bash -s -- --dry-run` 证据：`INSTALL.md` Claim：`clm_0008` supported 0.86
- `npx -y github:JuliusBrussee/caveman -- --list   # no clone needed` 证据：`INSTALL.md` Claim：`clm_0009` supported 0.86
- `git clone https://github.com/JuliusBrussee/caveman.git` 证据：`INSTALL.md` Claim：`clm_0010` supported 0.86
- `curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/src/rules/caveman-activate.md \` 证据：`INSTALL.md` Claim：`clm_0011` supported 0.86
- `npx -y github:JuliusBrussee/caveman -- --uninstall` 证据：`INSTALL.md` Claim：`clm_0012` supported 0.86
- `npm install -g @juliusbrussee/caveman-code` 证据：`README.md` Claim：`clm_0013` supported 0.86
- `curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bash -s -- --only openclaw` 证据：`README.md` Claim：`clm_0014` supported 0.86
- `npx -y github:JuliusBrussee/caveman -- --only openclaw` 证据：`README.md` Claim：`clm_0015` supported 0.86
- `npx skills@latest add JuliusBrussee/skills` 证据：`README.md` Claim：`clm_0016` 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：AI 研究者或研究型 Agent 构建者**（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：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_0005` 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：`plugins/caveman/skills/cavecrew/SKILL.md`, `plugins/caveman/skills/caveman/SKILL.md`, `plugins/caveman/skills/caveman-compress/SKILL.md`, `plugins/caveman/skills/caveman-stats/SKILL.md` 等 Claim：`clm_0006` 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：`plugins/caveman/skills/cavecrew/SKILL.md`, `plugins/caveman/skills/caveman/SKILL.md`, `plugins/caveman/skills/caveman-compress/SKILL.md`, `plugins/caveman/skills/caveman-stats/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`, `.claude-plugin/plugin.json`, `plugins/caveman/.codex-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：`INSTALL.md`, `README.md` Claim：`clm_0003` 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/marketplace.json`, `.claude-plugin/plugin.json`, `AGENTS.md`, `CLAUDE.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`, `.claude-plugin/plugin.json`, `plugins/caveman/.codex-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：`INSTALL.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：`INSTALL.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/marketplace.json`, `.claude-plugin/plugin.json`, `AGENTS.md`, `CLAUDE.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`, `.claude-plugin/plugin.json`, `INSTALL.md`, `README.md` 等
- **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：`CLAUDE.md`, `CONTRIBUTING.md`, `README.md`, `plugins/caveman/skills/caveman-compress/scripts/compress.py` 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_0017` 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`, `.claude-plugin/plugin.json`, `plugins/caveman/.codex-plugin/plugin.json` Claim：`clm_0018` 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：`INSTALL.md`, `README.md` Claim：`clm_0019` 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：`plugins/caveman/skills/cavecrew/SKILL.md`, `plugins/caveman/skills/caveman/SKILL.md`, `plugins/caveman/skills/caveman-compress/SKILL.md`, `plugins/caveman/skills/caveman-stats/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`, `.claude-plugin/plugin.json`, `plugins/caveman/.codex-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：`INSTALL.md`, `README.md` Claim：`clm_0003` supported 0.86

### 上下文规模

- 文件总数：107
- 重要文件覆盖：40/107
- 证据索引条目：73
- 角色 / Skill 条目：11

### 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 caveman 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 caveman 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 caveman 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 11 个角色 / Skill / 项目文档条目。

- **cavecrew**（skill）： 激活提示：当用户任务与“cavecrew”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：`plugins/caveman/skills/cavecrew/SKILL.md`
- **caveman-compress**（skill）： 激活提示：当用户任务与“caveman-compress”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：`plugins/caveman/skills/caveman-compress/SKILL.md`
- **caveman-stats**（skill）： 激活提示：当用户任务与“caveman-stats”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：`plugins/caveman/skills/caveman-stats/SKILL.md`
- **caveman**（skill）： 激活提示：当用户任务与“caveman”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：`plugins/caveman/skills/caveman/SKILL.md`
- **cavecrew**（skill）： 激活提示：当用户任务与“cavecrew”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/cavecrew/SKILL.md`
- **caveman-commit**（skill）： 激活提示：当用户任务与“caveman-commit”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/caveman-commit/SKILL.md`
- **caveman-compress**（skill）： 激活提示：当用户任务与“caveman-compress”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/caveman-compress/SKILL.md`
- **caveman-help**（skill）： 激活提示：当用户任务与“caveman-help”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/caveman-help/SKILL.md`
- **caveman-review**（skill）： 激活提示：当用户任务与“caveman-review”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/caveman-review/SKILL.md`
- **caveman-stats**（skill）： 激活提示：当用户任务与“caveman-stats”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/caveman-stats/SKILL.md`
- **caveman**（skill）： 激活提示：当用户任务与“caveman”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/caveman/SKILL.md`

## 证据索引

- 共索引 73 条证据。

- **CLAUDE.md — caveman**（documentation）：README = product front door. Non-technical people read it to decide if caveman worth install. Treat like UI copy. 证据：`CLAUDE.md`
- **Before / After**（documentation）：why use many token when few do trick 证据：`README.md`
- **Evals**（documentation）：Measures real token compression of caveman skills by running the same prompts through Claude Code under three conditions and comparing the generated output token counts. 证据：`evals/README.md`
- **cavecrew**（documentation）：Decision guide. When to delegate to caveman subagents instead of doing the work inline. 证据：`skills/cavecrew/README.md`
- **caveman-commit**（documentation）：Terse Conventional Commits. Why over what. 证据：`skills/caveman-commit/README.md`
- **What It Do**（documentation）：shrink memory file. save token every session. 证据：`skills/caveman-compress/README.md`
- **caveman-help**（documentation）：Quick-reference card. One shot, no mode change. 证据：`skills/caveman-help/README.md`
- **caveman-review**（documentation）：One-line PR comments. Location, problem, fix. No throat-clearing. 证据：`skills/caveman-review/README.md`
- **caveman-stats**（documentation）：Real session token receipts. No AI estimation. 证据：`skills/caveman-stats/README.md`
- **caveman**（documentation）：Talk like smart caveman. Same brain, fewer tokens. 证据：`skills/caveman/README.md`
- **Caveman Hooks**（documentation）：These hooks are bundled with the caveman plugin and activate automatically when the plugin is installed. No manual setup required. 证据：`src/hooks/README.md`
- **caveman-shrink**（documentation）：MCP middleware. Wrap any MCP server. Cut the prose. Keep the substance. 证据：`src/mcp-servers/caveman-shrink/README.md`
- **caveman — opencode plugin**（documentation）：Native opencode plugin. Mirrors the Claude Code hook architecture using opencode's session.created + tui.prompt.append lifecycle hooks. 证据：`src/plugins/opencode/README.md`
- **Package**（package_manifest）：{ "name": "caveman-installer", "version": "0.1.0", "description": "Caveman installer — detects your AI coding agents and installs caveman for each one.", "license": "MIT", "author": "Julius Brussee", "homepage": "https://github.com/JuliusBrussee/caveman", "repository": { "type": "git", "url": "git+https://github.com/JuliusBrussee/caveman.git" }, "bugs": { "url": "https://github.com/JuliusBrussee/caveman/issues" }, "bin": { "caveman": "./bin/install.js" }, "engines": { "node": " =18" }, "scripts": { "test": "node --test tests/installer/ .test.mjs" }, "files": "bin/", "src/", "agents/", "skills/", "plugins/", "commands/", "dist/caveman.skill", "README.md", "LICENSE" } 证据：`package.json`
- **Contributing to caveman**（documentation）：Thanks for considering a contribution. Caveman is a multi-agent skill that makes 30+ AI coding agents talk in compressed caveman-style prose. Most contributions fall into one of three buckets: 证据：`CONTRIBUTING.md`
- **Package**（package_manifest）：{ "type": "commonjs" } 证据：`src/hooks/package.json`
- **Package**（package_manifest）：{ "name": "caveman-shrink", "version": "0.1.0", "description": "MCP proxy that compresses prose fields tool descriptions, etc. using caveman rules. Same accuracy, fewer context tokens.", "license": "MIT", "homepage": "https://github.com/JuliusBrussee/caveman", "repository": { "type": "git", "url": "git+https://github.com/JuliusBrussee/caveman.git", "directory": "src/mcp-servers/caveman-shrink" }, "author": "Julius Brussee", "keywords": "mcp", "model-context-protocol", "claude", "caveman", "proxy", "tokens" , "main": "compress.js", "bin": { "caveman-shrink": "index.js" }, "files": "index.js", "compress.js", "README.md" } 证据：`src/mcp-servers/caveman-shrink/package.json`
- **Package**（package_manifest）：{ "name": "caveman-opencode-plugin", "version": "0.1.0", "description": "caveman plugin for opencode — terse mode tracking + slash commands", "type": "module", "main": "plugin.js", "private": true } 证据：`src/plugins/opencode/package.json`
- **When to use cavecrew vs alternatives**（skill_instruction）：Cavecrew = three subagent presets that emit caveman output. Same job as Anthropic defaults Explore , edit-style agents, reviewer ; difference is the tool-result they return is compressed, so main context shrinks per delegation. 证据：`plugins/caveman/skills/cavecrew/SKILL.md`
- **Caveman Compress**（skill_instruction）：Compress natural language files CLAUDE.md, todos, preferences into caveman-speak to reduce input tokens. Compressed version overwrites original. Human-readable backup saved as .original.md . 证据：`plugins/caveman/skills/caveman-compress/SKILL.md`
- **Skill**（skill_instruction）：This skill is delivered by hooks/caveman-stats.js read by hooks/caveman-mode-tracker.js on /caveman-stats . The model does not need to do anything when this skill fires — the hook returns decision: "block" with the formatted stats as the reason. The user sees the numbers immediately. 证据：`plugins/caveman/skills/caveman-stats/SKILL.md`
- **Persistence**（skill_instruction）：Respond terse like smart caveman. All technical substance stay. Only fluff die. 证据：`plugins/caveman/skills/caveman/SKILL.md`
- **When to use cavecrew vs alternatives**（skill_instruction）：Cavecrew = three subagent presets that emit caveman output. Same job as Anthropic defaults Explore , edit-style agents, reviewer ; difference is the tool-result they return is compressed, so main context shrinks per delegation. 证据：`skills/cavecrew/SKILL.md`
- **Rules**（skill_instruction）：Write commit messages terse and exact. Conventional Commits format. No fluff. Why over what. 证据：`skills/caveman-commit/SKILL.md`
- **Caveman Compress**（skill_instruction）：Compress natural language files CLAUDE.md, todos, preferences into caveman-speak to reduce input tokens. Compressed version overwrites original. Human-readable backup saved as .original.md . 证据：`skills/caveman-compress/SKILL.md`
- **Caveman Help**（skill_instruction）：Display this reference card when invoked. One-shot — do NOT change mode, write flag files, or persist anything. Output in caveman style. 证据：`skills/caveman-help/SKILL.md`
- **Rules**（skill_instruction）：Write code review comments terse and actionable. One line per finding. Location, problem, fix. No throat-clearing. 证据：`skills/caveman-review/SKILL.md`
- **Skill**（skill_instruction）：This skill is delivered by hooks/caveman-stats.js read by hooks/caveman-mode-tracker.js on /caveman-stats . The model does not need to do anything when this skill fires — the hook returns decision: "block" with the formatted stats as the reason. The user sees the numbers immediately. 证据：`skills/caveman-stats/SKILL.md`
- **Persistence**（skill_instruction）：Respond terse like smart caveman. All technical substance stay. Only fluff die. 证据：`skills/caveman/SKILL.md`
- **Agents**（documentation）：@./skills/caveman/SKILL.md @./skills/caveman-commit/SKILL.md @./skills/caveman-review/SKILL.md @./skills/caveman-compress/SKILL.md 证据：`AGENTS.md`
- **Gemini**（documentation）：@./skills/caveman/SKILL.md @./skills/caveman-commit/SKILL.md @./skills/caveman-review/SKILL.md @./skills/caveman-compress/SKILL.md 证据：`GEMINI.md`
- **Install caveman**（documentation）：One install. Works for every AI coding agent on your machine. 证据：`INSTALL.md`
- **Marketplace**（structured_config）：{ "$schema": "https://anthropic.com/claude-code/marketplace.schema.json", "name": "caveman", "description": "Ultra-compressed communication mode for Claude Code. Cuts ~75% of tokens while keeping full technical accuracy.", "owner": { "name": "Julius Brussee", "url": "https://github.com/JuliusBrussee" }, "plugins": { "name": "caveman", "description": "Talk like caveman. Cut ~75% tokens. Keep all technical accuracy.", "source": "./", "category": "productivity" } } 证据：`.claude-plugin/marketplace.json`
- **Plugin**（structured_config）：{ "name": "caveman", "description": "Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.", "author": { "name": "Julius Brussee", "url": "https://github.com/JuliusBrussee" }, "hooks": { "SessionStart": { "hooks": { "type": "command", "command": "node \"${CLAUDE PLUGIN ROOT}/src/hooks/caveman-activate.js\"", "timeout": 5, "statusMessage": "Loading caveman mode..." } } , "UserPromptSubmit": { "hooks": { "type": "command", "command": "node \"${CLAUDE PLUGIN ROOT}/src/hooks/caveman-mode-tracker.js\"", "timeout": 5, "statusMessage": "Tracking caveman mode..." } } } } 证据：`.claude-plugin/plugin.json`
- **Plugin**（structured_config）：{ "name": "caveman", "version": "0.1.0", "description": "Ultra-compressed communication mode. Cut filler. Keep technical accuracy.", "author": { "name": "Julius Brussee", "url": "https://github.com/JuliusBrussee" }, "homepage": "https://github.com/JuliusBrussee/caveman", "repository": "https://github.com/JuliusBrussee/caveman", "license": "MIT", "keywords": "productivity", "communication", "brevity", "writing" , "skills": "./skills/", "interface": { "displayName": "Caveman", "shortDescription": "Talk like caveman. Cut filler. Keep technical accuracy.", "longDescription": "Ultra-compressed communication mode for Codex. Use fewer words. Keep exact technical substance.", "developerName": "Juli… 证据：`plugins/caveman/.codex-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`
- **Caveman Commit**（documentation）：Generate a commit message for the current staged changes. 证据：`src/plugins/opencode/commands/caveman-commit.md`
- **Caveman Compress**（documentation）：Run the caveman-compress skill against the given filepath. The skill rewrites prose into terse caveman style — drops articles, filler, hedging — while preserving code blocks, inline code, URLs, file paths, commands, and markdown structure exactly. Original is backed up as .original.md before overwrite. 证据：`src/plugins/opencode/commands/caveman-compress.md`
- **Caveman Activate**（documentation）：Respond terse like smart caveman. All technical substance stay. Only fluff die. 证据：`src/rules/caveman-activate.md`
- **Init**（source_file）：all = "cli", "compress", "detect", "validate" version = "1.0.0" 证据：`plugins/caveman/skills/caveman-compress/scripts/__init__.py`
- **YAML frontmatter: starts at file start with --- on its own line, ends with --- on its own line.**（source_file）：OUTER FENCE REGEX = re.compile YAML frontmatter: starts at file start with --- on its own line, ends with --- on its own line. Captures the entire block including delimiters and trailing newline and the body after. FRONTMATTER REGEX = re.compile def split frontmatter text: str ⋮---- """Split YAML frontmatter from body. Returns frontmatter, body . Memory files and many other markdown docs start with a YAML frontmatter block delimited by --- lines. The compression LLM has a habit of stripping or rewriting these despite preserve-structure rules in the prompt — so we surgically remove the frontmatter before compression and prepend it back verbatim to the output. Files without frontmatter pass t… 证据：`plugins/caveman/skills/caveman-compress/scripts/compress.py`
- **Init**（source_file）：all = "cli", "compress", "detect", "validate" version = "1.0.0" 证据：`skills/caveman-compress/scripts/__init__.py`
- **YAML frontmatter: starts at file start with --- on its own line, ends with --- on its own line.**（source_file）：OUTER FENCE REGEX = re.compile YAML frontmatter: starts at file start with --- on its own line, ends with --- on its own line. Captures the entire block including delimiters and trailing newline and the body after. FRONTMATTER REGEX = re.compile def split frontmatter text: str ⋮---- """Split YAML frontmatter from body. Returns frontmatter, body . Memory files and many other markdown docs start with a YAML frontmatter block delimited by --- lines. The compression LLM has a habit of stripping or rewriting these despite preserve-structure rules in the prompt — so we surgically remove the frontmatter before compression and prepend it back verbatim to the output. Files without frontmatter pass t… 证据：`skills/caveman-compress/scripts/compress.py`
- **Compress**（source_file）：/ ^ \n + /g, // inline code /\bhttps?:\/\/\S+/gi, // URLs /\b \w.- \/\\ \w.\/\\\- +/g, // paths with / or \ /\b A-Z A-Za-z0-9 ?: A-Z A-Za-z0-9 +\b/g, // CONST CASE /\b\w+\.\w+ ?:\.\w+ \ \ ?/g, // dotted.method or pkg.fn / A-Za-z A-Za-z0-9 \s \ ^ \ /g, // function calls /\b\d+\.\d+\.\d+\b/g, // version numbers ⋮---- function withProtectedSegments text, transform ⋮---- // Replace every protected match with a sentinel, transform the rest, then // splice the originals back in. ⋮---- function compressProse text ⋮---- // Collapse repeated whitespace introduced by removals. ⋮---- // Capitalize the first letter of each sentence we may have left lowercase. ⋮---- function compress text, opts // Walk… 证据：`src/mcp-servers/caveman-shrink/compress.js`
- **Scope**（documentation）：Caveman-ultra. Drop articles/filler. Code/paths exact, backticked. No narration. 证据：`agents/cavecrew-builder.md`
- **Job**（documentation）：Caveman-ultra. Drop articles/filler/hedging. Code/symbols/paths exact, backticked. Lead with answer. 证据：`agents/cavecrew-investigator.md`
- **Severity**（documentation）：Caveman-ultra. Findings only. No "looks good", no "I'd suggest", no preamble. 证据：`agents/cavecrew-reviewer.md`
- **Scope**（documentation）：Caveman-ultra. Drop articles/filler. Code/paths exact, backticked. No narration. 证据：`plugins/caveman/agents/cavecrew-builder.md`
- **Job**（documentation）：Caveman-ultra. Drop articles/filler/hedging. Code/symbols/paths exact, backticked. Lead with answer. 证据：`plugins/caveman/agents/cavecrew-investigator.md`
- **Severity**（documentation）：Caveman-ultra. Findings only. No "looks good", no "I'd suggest", no preamble. 证据：`plugins/caveman/agents/cavecrew-reviewer.md`
- **Security**（documentation）：caveman-compress receives a Snyk High Risk rating due to static analysis heuristics. This document explains what the skill does and does not do. 证据：`skills/caveman-compress/SECURITY.md`
- **Caveman Help**（documentation）：Show the caveman quick-reference card. 证据：`src/plugins/opencode/commands/caveman-help.md`
- **Caveman Review**（documentation）：Review the current diff or files: $ARGUMENTS . 证据：`src/plugins/opencode/commands/caveman-review.md`
- **Caveman Stats**（documentation）：Show caveman stats — total tokens saved, sessions, average compression ratio. 证据：`src/plugins/opencode/commands/caveman-stats.md`
- **Caveman**（documentation）：If no level given, use full. If "off", deactivate. 证据：`src/plugins/opencode/commands/caveman.md`
- **Caveman mode always on**（documentation）：Respond terse like smart caveman. All technical substance stay. Only fluff die. 证据：`src/rules/caveman-openclaw-bootstrap.md`
- **Hooks**（structured_config）：{ "hooks": { "SessionStart": { "matcher": "startup resume", "hooks": { "type": "command", "command": "echo 'CAVEMAN MODE ACTIVE. Rules: Drop articles/filler/pleasantries/hedging. Fragments OK. Short synonyms. Pattern: thing action reason . next step . Not: Sure! I would be happy to help you with that. Yes: Bug in auth middleware. Fix: Code/commits/security: write normal. User says stop caveman or normal mode to deactivate.'", "timeout": 5, "statusMessage": "Loading caveman mode" } } } } 证据：`.codex/hooks.json`
- **Gemini Extension**（structured_config）：{ "name": "caveman", "description": "Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.", "version": "1.0.1", "contextFileName": "GEMINI.md" } 证据：`gemini-extension.json`
- **Skills Lock**（structured_config）：{ "version": 1, "skills": { "cavecrew": { "source": "JuliusBrussee/caveman", "sourceType": "github", "skillPath": "skills/cavecrew/SKILL.md", "computedHash": "06d45a7308d8603365313decf400020106b985cb5ce500ee169bfba9c71dd147" } } } 证据：`skills-lock.json`
- **Results**（structured_config）：{ "metadata": { "generated at": "2026-04-08T22:01:24.027359+00:00", "claude cli version": "2.1.97 Claude Code ", "model": "claude-opus-4-6", "n prompts": 10, "terse prefix": "Answer concisely." }, "prompts": "Why does my React component re-render every time the parent updates?", "Explain database connection pooling.", "What's the difference between TCP and UDP?", "How do I fix a memory leak in a long-running Node.js process?", "What does the SQL EXPLAIN command tell me?", "How does a hash table handle collisions?", "Why am I getting CORS errors in my browser console?", "What's the point of using a debouncer on a search input?", "How does git rebase differ from git merge?", "When should I us… 证据：`evals/snapshots/results.json`
- 其余 13 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：`CLAUDE.md`, `README.md`, `evals/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：`CLAUDE.md`, `README.md`, `evals/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。

- **项目概览与压缩原理**：importance `high`
  - source_paths: README.md, skills/caveman/SKILL.md, src/rules/caveman-activate.md, src/rules/caveman-openclaw-bootstrap.md, CLAUDE.md
- **多代理安装与分发系统**：importance `high`
  - source_paths: bin/install.js, bin/lib/openclaw.js, bin/lib/opencode-agent.js, bin/lib/settings.js, install.sh
- **技能、子代理与命令套件**：importance `high`
  - source_paths: skills/caveman/SKILL.md, skills/caveman/README.md, skills/caveman-commit/SKILL.md, skills/caveman-review/SKILL.md, skills/caveman-stats/SKILL.md
- **钩子架构、MCP 中间件与安全模型**：importance `high`
  - source_paths: src/hooks/caveman-activate.js, src/hooks/caveman-mode-tracker.js, src/hooks/caveman-config.js, src/hooks/caveman-stats.js, src/hooks/caveman-statusline.sh

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `25d22f864ad68cc447a4cb93aefde918aa4aec9f`
- inspected_files: `README.md`, `package.json`, `docs/install-windows.md`, `src/tools/caveman-init.js`, `src/hooks/README.md`, `src/hooks/package.json`, `src/hooks/caveman-activate.js`, `src/hooks/caveman-mode-tracker.js`, `src/hooks/caveman-stats.js`, `src/hooks/caveman-config.js`, `src/rules/caveman-activate.md`, `src/rules/caveman-openclaw-bootstrap.md`, `src/plugins/opencode/README.md`, `src/plugins/opencode/package.json`, `src/plugins/opencode/plugin.js`, `src/plugins/opencode/commands/caveman-compress.md`, `src/plugins/opencode/commands/caveman-commit.md`, `src/plugins/opencode/commands/caveman-review.md`, `src/plugins/opencode/commands/caveman-stats.md`, `src/plugins/opencode/commands/caveman-help.md`

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: 来源证据：caveman does not work when installing for opencode

- Trigger: GitHub 社区证据显示该项目存在一个安装相关的待验证问题：caveman does not work when installing for opencode
- Why it matters: 可能阻塞安装或首次运行。
- Evidence: community_evidence:github | https://github.com/JuliusBrussee/caveman/issues/482 | 来源讨论提到 node 相关条件，需在安装/试用前复核。
- 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: 来源证据：npx install for opencode incomplete — no such file or directory src\plugins\opencode\commands\caveman-compress.md

- Trigger: GitHub 社区证据显示该项目存在一个安装相关的待验证问题：npx install for opencode incomplete — no such file or directory src\plugins\opencode\commands\caveman-compress.md
- Why it matters: 可能阻塞安装或首次运行。
- Evidence: community_evidence:github | https://github.com/JuliusBrussee/caveman/issues/464 | 来源类型 github_issue 暴露的待验证使用条件。
- 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: 失败模式：security_permissions: Security Scan Flags “Caution” Verdict for the *caveman* Skill (High‑risk Exfiltration & Mediu...

- Trigger: Developers should check this security_permissions risk before relying on the project: Security Scan Flags “Caution” Verdict for the *caveman* Skill (High‑risk Exfiltration & Medium‑risk Path Traversal)
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: Security Scan Flags “Caution” Verdict for the *caveman* Skill (High‑risk Exfiltration & Medium‑risk Path Traversal). Context: Observed during installation or first-run setup.
- Why it matters: Developers may expose sensitive permissions or credentials: Security Scan Flags “Caution” Verdict for the *caveman* Skill (High‑risk Exfiltration & Medium‑risk Path Traversal)
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/528 | Security Scan Flags “Caution” Verdict for the *caveman* Skill (High‑risk Exfiltration & Medium‑risk Path Traversal)
- 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 4: 来源证据：Can't find or activate /caveman in IBM Bob

- Trigger: GitHub 社区证据显示该项目存在一个安全/权限相关的待验证问题：Can't find or activate /caveman in IBM Bob
- Why it matters: 可能影响授权、密钥配置或安全边界。
- Evidence: community_evidence:github | https://github.com/JuliusBrussee/caveman/issues/523 | 来源类型 github_issue 暴露的待验证使用条件。
- 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: 来源证据：Portuguese (PT-BR) adaptation: Troglodita

- Trigger: GitHub 社区证据显示该项目存在一个安全/权限相关的待验证问题：Portuguese (PT-BR) adaptation: Troglodita
- Why it matters: 可能影响授权、密钥配置或安全边界。
- Evidence: community_evidence:github | https://github.com/JuliusBrussee/caveman/issues/457 | 来源类型 github_issue 暴露的待验证使用条件。
- 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: 失败模式：installation: Can't find or activate /caveman in IBM Bob

- Trigger: Developers should check this installation risk before relying on the project: Can't find or activate /caveman in IBM Bob
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: Can't find or activate /caveman in IBM Bob. Context: Observed during installation or first-run setup.
- Why it matters: Developers may fail before the first successful local run: Can't find or activate /caveman in IBM Bob
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/523 | Can't find or activate /caveman in IBM Bob
- 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: 失败模式：installation: Portuguese (PT-BR) adaptation: Troglodita

- Trigger: Developers should check this installation risk before relying on the project: Portuguese (PT-BR) adaptation: Troglodita
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: Portuguese (PT-BR) adaptation: Troglodita. Context: Observed during installation or first-run setup.
- Why it matters: Developers may fail before the first successful local run: Portuguese (PT-BR) adaptation: Troglodita
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/457 | Portuguese (PT-BR) adaptation: Troglodita
- 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 8: 失败模式：installation: ask about project

- Trigger: Developers should check this installation risk before relying on the project: ask about project
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: ask about project. Context: Observed during installation or first-run setup.
- Why it matters: Developers may fail before the first successful local run: ask about project
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/522 | ask about project
- 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 9: 失败模式：installation: caveman does not work when installing for opencode

- Trigger: Developers should check this installation risk before relying on the project: caveman does not work when installing for opencode
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: caveman does not work when installing for opencode. Context: Observed when using node
- Why it matters: Developers may fail before the first successful local run: caveman does not work when installing for opencode
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/482 | caveman does not work when installing for opencode
- 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 10: 失败模式：installation: caveman-shrink MCP registered with no upstream → fails on every load

- Trigger: Developers should check this installation risk before relying on the project: caveman-shrink MCP registered with no upstream → fails on every load
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: caveman-shrink MCP registered with no upstream → fails on every load. Context: Observed when using node
- Why it matters: Developers may fail before the first successful local run: caveman-shrink MCP registered with no upstream → fails on every load
- Evidence: failure_mode_cluster:github_issue | https://github.com/JuliusBrussee/caveman/issues/474 | caveman-shrink MCP registered with no upstream → fails on every load
- 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。