explore-harness
Wrapper agent that caches explore results and constrains the scope of exploration.
explore-harness wraps the explore agent instead of calling it directly. It caches exploration results and restricts the scope of each search, so when several agents keep exploring the same repository it cuts the redundant work, and keeping scope explicit holds the noise down.
Role
- Cache
exploreresults within a session to avoid repeated exploration cost - Constrain exploration to specific directories, file patterns, or symbol sets
- Manage per-hypothesis exploration results separately during multi-hypothesis investigations
- Coordinate parallel explorations when
debuggerorplannerneeds to explore multiple paths simultaneously
When invoked
| Situation | How it's triggered |
|---|---|
When debugger is tracking multiple hypotheses in parallel | Delegated from debugger |
| When repeated exploration of a large repository must be avoided | Orchestrator request |
| When exploration scope must be explicitly narrowed to a specific module | Direct request |
Difference from explore
explore on its own searches a repository broadly. explore-harness wraps it and layers on two things: scope limiting and result reuse. For a straightforward lookup, plain explore is plenty. Where the harness earns its keep is the messy investigation that needs the same ground covered repeatedly, or several paths explored in parallel.
Limits
- Does not create, modify, or delete any files
- Does not make architectural judgments or modification recommendations
- Cache is valid only within the current session