用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/ZzjNoMercy/PuddingClaw --skill knowledge-search命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | knowledge-search |
| description | Retrieve answers from local document knowledge bases. |
| toolsets | ["knowledge_analysis","feishu_bitable"] |
For document questions, call llamaindex_knowledge_query after activating this
Skill. Reading this file only activates the toolset; it does not perform a
retrieval. Markdown LLM Wiki and GBrain are separate knowledge paths and do not
replace this document-index query. Do not use spreadsheet analysis for PDF or
Markdown content.
Treat one successful retrieval as the authoritative result set for the current
question. Do not follow it with exploratory grep, glob, or ls. The tool
reports text and image hit counts and returns exact /knowledge/... paths. If
the answer needs more source text, call read_file only on the exact linked
Markdown virtual path returned by the retrieval. If a returned image matters,
use its exact virtual path with the image-analysis workflow; never rediscover it
by scanning the knowledge tree.
For a registered Feishu Bitable source, call feishu_bitable_list_sources when
the exact source ID is unknown, then call feishu_bitable_describe to read
its live schema and then feishu_bitable_query for one bounded page. These
tools do not query arbitrary links or persist returned record values.