explore
Read-only codebase discovery agent that maps files, symbols, entry points, and dependency relationships.
The explore agent answers one question: where does X live? It runs first, before any other agent touches the code, fanning out broad-to-narrow parallel searches across the repo and handing back a structured report of file paths, symbol locations, and call-graph relationships. And because it only ever reads, it's safe to drop into any point of a workflow with zero side effects.
Role
- Map files, directories, and module boundaries relevant to a query
- Locate symbols (functions, classes, types) and trace their cross-file relationships
- Identify entry points, dependency chains, and import graphs
- Cross-validate findings across Glob, Grep, AST search, and LSP to eliminate false positives
When invoked
| Situation | How it's triggered |
|---|---|
At the start of $autopilot, $deep-interview, and $deepsearch | Automatic |
| "Find where X lives", "which files touch Y" | Direct request |
When planner or executor needs concrete file paths | Parent agent request |
| During debugging when the source location of a symptom is unknown | Delegated from debugger |
Inputs
- Repo root path
- A search query or scope hint (symbol name, feature area, file pattern, or free-form description)
- Optional: prior context such as error messages or partially known file paths
Outputs
- An in-session report listing absolute file paths, matched symbols, and relationship summaries
- For deep-search runs, findings may be persisted to
.omx/research/<topic>.mdfor downstream agents to reference - A
next_stepssection indicating which agent should act on the findings
Limits
- Does not create, modify, or delete any file under any circumstance
- Does not run tests, execute build commands, or assess code quality
- Does not make architectural recommendations — that is the
architectagent's role