| name | code-query |
| description | Search the codebase and Obsidian vault for relevant implementation and project context. Use when the user asks how something works, where it is implemented, or wants related notes alongside code results. |
Code Query
Search codebase and vault documents for relevant information.
Usage
/code-query <search-term>
Where $ARGUMENTS is the search term or question to investigate.
Actions
- Invoke the
vault skill
- Search the codebase: for unison projects use the unison MCP server; if
.codegraph/ exists use the CodeGraph routing below; otherwise search files using Grep/Glob
- Search vault notes for related context and documentation
- Return consolidated results with:
- Relevant code locations (file:line)
- Related vault notes
- Summary of findings
CodeGraph routing
When .codegraph/ exists, route by question shape:
| Question | Tool |
|---|
| "Where is X defined?" / "Find symbol named X" | codegraph_search |
| "What calls function Y?" | codegraph_callers |
| "What does Y call?" | codegraph_callees |
| "How does X reach/become Y? / trace the flow" | codegraph_trace (one call = the whole path, incl. callback/React/JSX dynamic hops) |
| "What would break if I changed Z?" | codegraph_impact |
| "Show me Y's signature / source / docstring" | codegraph_node |
| "Give me focused context for a task/area" | codegraph_context |
| "See several related symbols' source at once" | codegraph_explore |
| "What files exist under path/" | codegraph_files |
| "Is the index healthy?" | codegraph_status |
Rules of thumb:
- Answer directly — don't delegate exploration. For "how does X work" questions:
codegraph_context first, then ONE codegraph_explore for the source it surfaces. For flows, codegraph_trace from→to returns the whole path in one call — don't rebuild it with codegraph_search + codegraph_callers. CodeGraph IS the pre-built index; grep+read loops repeat its work at higher cost.
- Trust indexed results — don't re-grep them unless exhaustive textual coverage matters (strings, comments, config, generated files — those are
rg's surfaces).
- Don't grep first for a symbol by name;
codegraph_search returns kind + location + signature in one call.
- Don't chain search + node for context (
codegraph_context is one call) and don't loop codegraph_node over many symbols (one codegraph_explore is cheaper).
- Hybrid rule: for broad "find X" searches, CodeGraph first, then one targeted
rg pass when the term also appears as text or results look sparse.
- Staleness banner: files listed as pending re-index → Read those files; everything else is fresh.
- If the MCP server says "not initialized", ask before running
codegraph init -i; if declined, locate-then-read with rg.
Examples
/code-query authentication flow
/code-query how errors are handled
/code-query database connection
