auto-research - Iteratively improve local artifacts with scored changes
Runs a bounded hill-climbing loop that proposes, scores, keeps, or reverts changes to a local artifact using an independent checker.
Tags
Updated: 2026-10-04Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Iterate a local artifact
- Draft a governance program
- Propose one specific change
- Score changes independently
- Keep higher-scoring versions
- Revert worse versions
- Track iteration results
- Report iteration status
- Repair stalled loops
Inputs
- Local artifact
- Quality rubric
- Iteration budget
- Governance program
- User confirmation
Outputs
- Improved artifact copy
- Original artifact preserved
- Iteration history
- Research status file
- Per-iteration status messages
- Best-version reference
Requirements
- Workspace-scoped local filesystem
- Write access to the workspace
- Git reset support
- loop-checker agent
- Haiku model support