agentskills.codes

Install

mkdir -p .claude/skills/search-v1ceversaa && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/13612" && unzip -o skill.zip -d .claude/skills/search-v1ceversaa && rm skill.zip

Installs to .claude/skills/search-v1ceversaa

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Use when the user wants to find existing content in the repo — "有没有关于 X 的笔记"、"我之前读过哪些 Y"、"把和 Z 相关的内容列出来"、"find notes on"、"search for". This is a read-only skill: it never writes files, never modifies overviews, never updates indexes. Use even when the user does not say /search.
278 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)

About this skill

/search

When To Use

  • The user wants to retrieve or discover existing notes, topics, syntheses, conferences, or local ideas.
  • The user is asking "what do we already have" before deciding what to read or synthesize.

Hard Boundaries (Read-Only)

  • Does not write any markdown files
  • Does not modify overview.md
  • Does not update indexes
  • Does not append to ops/log.md
  • Does not archive PDFs

Contrast with similar skills:

  • /search: find and list existing content, explain relevance
  • /synthesize: compare content and produce a new written output
  • /discuss: use retrieved content to push a research idea forward

Workflow

  1. Extract key terms from the user's query.
  2. Run query_context.py with the query:
    python3 .codex/skills/maintain/scripts/query_context.py \
      "<user query terms>" \
      --limit 8
    
    Add --include-private if the user may be looking for local ideas or drafts.
  3. Present results in ranked order. For each result, provide:
    • Path and title
    • 1–2 sentences explaining why it's relevant to the query
    • Current status if informative (e.g., drafting vs complete)
  4. Recommend a next step for the most relevant results:
    • Note is complete and the user wants to review → /refresh
    • Note is missing and user wants to read the paper → /read or /intake
    • Multiple relevant notes suggest a comparison → /synthesize
    • Results spark a research idea → /discuss

Inputs Expected

  • Free-text query (topic, paper title fragment, concept, author name, venue)

Scripts

  • ../maintain/scripts/query_context.py — full invocation example in _shared/script-registry.md

Done Criteria

  • Ranked results are presented with relevance explanations
  • A next step is suggested for the most relevant hits

Search skills

Search the agent skills registry