| name | library-search |
| description | Search gegeewu-library archives using three-layer search (JSON index + Obsidian REST API full-text + structured query). 用户说"搜索知识库"、"查找归档"、"library search"时触发。 |
| user-invocable | true |
| allowed-tools | Bash Read Grep |
library-search
Use this skill before any new intake or to search existing archives.
Goal
- Prevent duplicate archives
- Return the best existing match and the next action
- Support multi-dimensional search (exact, full-text, structured)
Search Modes
Mode A: Duplicate Check (intake internal)
- Execute: Layer 1 + Layer 2
- Return: hit/miss + recommended action (update/create new)
Mode B: User Search
- Execute: Layer 1 + Layer 2 + Layer 3 (as needed)
- Return: matching entries list + summary info
Three-Layer Search
Layer 1: JSON Exact Match
- File:
reference/library-index.json
- Match against: name, slug, aliases (exact or fuzzy)
- Optional filter: type, category, tags
Layer 2: Obsidian REST API Full-Text
- API:
POST https://127.0.0.1:27124/search/simple/?query=<keyword>
- Header:
Authorization: Bearer ${OBSIDIAN_REST_API_KEY}
- TLS: skip cert verification
Layer 3: JSON Structured Filtering
- Filter by type / category / tags from
library-index.json
- Examples:
- "All AI agent projects" → type=project, category=ai/agent
- "CLI tools" → tags contains "cli"
- "Recent archives" → sort by updated_at desc
Search Routing
- Duplicate check: Layer 1 + Layer 2
- Exact lookup ("is opencli archived?"): Layer 1 only
- Fuzzy search ("CLI related"): Layer 1 tags + Layer 2 full-text
- Category browse ("all AI agent"): Layer 3 structured
- Broad search ("tools for posting to XHS"): Layer 1 + Layer 2 combined
Expected Output
- Type, match status (exact / close / none)
- Existing name, path, category and tags
- Recommended action: update existing / create new
Hard Rules
- JSON index alone is not enough; check Obsidian too
- If the two disagree, report drift
- If REST API is unavailable: stop and report error
- Do not falsely report search results
