AGENT SKILL

query

Query engram's local knowledge graph for structural context — function calls, imports, type relationships, mistake history, ADRs. Use when the user asks "how does X work in this project", "what calls Y", "where is Z used", or any structural question that doesn't need file content.

License 许可
Apache-2.0
所属仓库
NickCirv/engram
最近更新
2026-10-21
Stars
★ 143
Forks
15

01 前言 Front Matter

description
Query engram's local knowledge graph for structural context — function calls, imports, type relationships, mistake history, ADRs. Use when the user asks "how does X work in this project", "what calls Y", "where is Z used", or any structural question that doesn't need file content.

03 SKILL.md 正文

engram query

Query engram's knowledge graph instead of reading files directly. Saves tokens when a structural answer suffices.

The argument is a natural-language question. Run:

engram query "$ARGUMENTS" -p $CLAUDE_PROJECT_DIR

The graph returns a token-budgeted answer with:

  • Relevant nodes (functions, files, concepts)
  • Edges (calls, imports, decided-for relationships)
  • Mistake hits if any (these surface with ⚠️ at the top — read carefully, they represent past failure modes)

If the answer doesn't fully address the question, fall back to reading specific files mentioned in the result.

For a connection between two specific concepts use engram path <source> <target>. For most-connected entities use engram gods.

Example invocations

User: "How does authentication work in this project?" You: Run engram query "how does authentication work" -p $CLAUDE_PROJECT_DIR. Read the structured response. Cite the relevant nodes (function names, file paths). If the answer mentions specific files, those are good candidates for a follow-up Read — but only if the structural view leaves a question.

User: "What calls validateToken?" You: Run engram query "what calls validateToken" -p $CLAUDE_PROJECT_DIR. The graph returns the inbound call sites without you ever Reading those files.

User: "Where is the rate limiter implemented?" You: Run engram query "rate limiter implementation" -p $CLAUDE_PROJECT_DIR. If the answer points at one file, you can Read that file confidently. If it points at three files, ask the user which path matters.

User: "Trace from userController.create to the database write" You: This is a path query, not a free-text query. Use engram path userController.create database.write -p $CLAUDE_PROJECT_DIR. Returns the shortest call chain.

User: "Show me the most-connected files in this codebase" You: Use engram gods -p $CLAUDE_PROJECT_DIR --top 10. Returns the entities with the most graph degree. Useful for "where do I start reading."

When NOT to fire this skill

  • The user explicitly asks to read a specific file. They want the file content, not a summary. Use the Read tool.
  • The user is debugging a runtime error and needs the exact stack trace. The graph is structural; runtime data lives in logs.
  • The graph is empty (engram stats shows 0 nodes) — the project hasn't been indexed yet. Recommend engram init . first.

04 同仓库更多 Skills