《毛选》1-7卷文本查询

触发：当用户要检索《毛泽东选集》全文、按卷或文章定位原文、按标题/关键词找内容时使用。此 skill 适用于 OpenClaw，本地默认使用结构化检索与关键词检索；只有在配置文件显式开启后，才使用向量召回和重排组成的混合检索。

Security Scan

The skill's files and instructions largely match a local document-search capability, but it conditionally calls external embedding/rerank APIs and can inherit a global API key configuration — behavior that is disproportionate unless explicitly enabled and inspected.

Purpose & Capability

Name/description, included data/ corpus and the search/build_index scripts align with a local retrieval/search skill for 《毛泽东选集》. The code and docs describe building a local SQLite/FTS index and optional vector indexes — these are expected for the stated purpose.

Instruction Scope

SKILL.md limits external calls to an opt-in RAG mode (rag.enabled) and documents local-only defaults. However the runtime instructions include commands to 'test-model', enable 'rag.enabled' and set API keys; those commands will cause network calls outside of the local corpus if enabled. The instructions otherwise stay within the stated retrieval scope and reference only the local data/ directory.

Install Mechanism

No install spec is declared (instruction-only install). The skill includes Python scripts (build_index.py, search.py, etc.) and reads local Markdown files — no dubious remote download/install behavior is declared.

Credentials

The manifest declares no required env vars, but SKILL.md recommends MAO_SKILL_API_KEY for embeddings and documents that embedding/rerank configuration will inherit rag.api.base_url and rag.api.api_key_env by default. Automatically inheriting a global rag.* API key setting is disproportionate: it risks the skill using platform/global credentials without a clear explicit opt-in. Also the SKILL.md example default base_url (https://api.siliconflow.cn/v1) is a third‑party endpoint; if enabled it would send text/embeddings to an external service. Require explicit, per-skill credential configuration rather than implicit inheritance.

Persistence & Privilege

always is false and there is no install routine that requests permanent system-wide presence or writes to other skill configs. The skill stores and reads its own data/ and config/search.json only.

What to consider before installing

This skill appears to be a local search tool over the included corpus and is coherent with that purpose — but pay attention before enabling the hybrid/vector (RAG) mode. By default the SKILL.md states it operates locally, but if you turn on rag.enabled the scripts will call embedding/reranker APIs and will look for an API key. Two things to check before installing or using: - Keep rag.enabled false unless you intentionally want remote embedding/reranking. With rag.enabled=false the skill uses local FTS/lexical search only. - If you do enable rag, set a dedicated environment variable (e.g., MAO_SKILL_API_KEY) and a dedicated base_url for only this skill. Do NOT let the skill inherit a platform-wide rag.api.api_key_env value — that could expose a global/API key used by other integrations. Confirm config/search.json is edited to point to a credential you control. - Inspect the scripts (scripts/search.py, build_index.py, common.py) to confirm precisely what is sent to the remote endpoint (full text, excerpts, or just hashes). If you are privacy-sensitive, keep RAG disabled or run embedding with a local model/provider you trust. - Note the skill includes ~400 local Markdown files (large corpus). Confirm you are allowed to host/use this corpus and that local storage use is acceptable. If you want, I can (a) scan the scripts to list exact network calls/endpoints and what payloads they transmit, or (b) suggest exact config edits to ensure the skill never uses external API keys unless you explicitly set them.

毛选检索

Overview

这个 skill 为 OpenClaw 提供《毛泽东选集》的本地知识库能力，包含 2 个部分：

• scripts/build_index.py：从 Markdown 建立 SQLite FTS 索引，并按配置决定是否生成向量索引
• scripts/search.py：按卷、篇、标题、关键词或混合检索查询文章与段落

默认模式不依赖外部 API，只使用本地结构化检索与全文检索。只有 config/search.json 中显式开启 rag.enabled 时，才会使用 embedding；只有 rag.rerank.enabled 也打开时，才会继续重排。

默认模型与平台约定：

• 嵌入模型：BAAI/bge-m3
• 重排模型：BAAI/bge-reranker-v2-m3
• 嵌入分批默认按 64 条请求，避免常见平台的单批上限
• 如果用户没有单独为 embedding / rerank 配置 base_url 和 api_key_env，则默认继承 rag.api.base_url 与 rag.api.api_key_env

何时使用

在这些场景调用本 skill：

• 用户要找《毛选》第几卷、某篇文章或某个主题出现在哪里
• 用户给出标题、别名或关键词，希望返回对应文章或相关段落
• 用户要求开启混合检索，并提供可用的 embedding / rerank API 配置

以下情况不要直接调用本 skill：

• 用户只是在讨论观点，不需要定位《毛选》原文
• 当前工作已经有准确文件路径和文本片段，不需要再次建索引或检索

工作流

1. 建立索引

直接运行：

python scripts/build_index.py

默认会建立：

• 文档级索引：按卷、篇、标题、别名、全文找文章
• 段落级索引：按关键词找命中片段

索引工具默认直接扫描 data/ 目录下的 Markdown 文件。

如果 config/search.json 里开启了 rag.enabled 并配置好 embedding，则会同时生成向量索引。

2. 配置

API key 建议通过环境变量 MAO_SKILL_API_KEY 配置。

使用配置管理脚本：

# 查看当前配置
python scripts/config.py show

# 修改配置
python scripts/config.py set rag.api.base_url "https://api.siliconflow.cn/v1"
python scripts/config.py set rag.api.api_key_env "MAO_SKILL_API_KEY"
python scripts/config.py set rag.enabled true
python scripts/config.py set chunk_size 1024
python scripts/config.py set chunk_overlap 100

其他模型参数（如 embedding.model、embedding.batch_size、rerank.model）直接编辑 config/search.json。

3. 查询

常用查询方式：

按卷列文章：

python scripts/search.py catalog --volume 第一卷

按卷和篇直接定位文章：

python scripts/search.py show --volume 1 --chapter 3

按标题找文章：

python scripts/search.py show --title 实践论

按关键词搜段落：

python scripts/search.py search "调查研究"

显式开启混合检索：

python scripts/search.py search "统一战线" --mode hybrid

测试模型连通性：

python scripts/search.py test-model --target embedding
python scripts/search.py test-model --target rerank

如果没有配置 rag.api.base_url，或没有设置环境变量 MAO_SKILL_API_KEY，命令会直接提示用户先设置。

卷、篇、标题、关键词混合过滤：

python scripts/search.py search "调查研究" --volume 1 --chapter 7 --title 本本主义

输出要求

调用本 skill 时，返回结果必须优先给出可核对来源：

• 卷次
• 篇次
• 文章标题
• 日期（如果有）
• 命中的段落摘要
• 源文件路径或文档 ID
• 检索方式：catalog、lexical、lexical-like、hybrid、hybrid-rerank

如果没有命中，不要臆造答案。应明确说明：

• 没有在当前语料中检到
• 是标题未命中，还是关键词未命中
• 如果用户允许，可以建议补充别名、整理 metadata 或开启混合检索

参考资料

• 数据格式：references/corpus-format.md
• 检索规则：references/retrieval-rules.md
• 输出结构：references/output-schema.md