03 · Improve
Improve
Runs the diagnose, mutate, evaluate loop and ships verified fixes
Takes diagnostics results and known issues, generates and applies the mutation to your target platform, local code, GitHub PR, or Paperclip, validates it against the Evaluator, and verifies it works after the update. Loops until the change beats baseline.
Mutation Agent's discipline is execution with validation. It does not invent strategies, it implements them. It does not diagnose, it consumes diagnostics. It loops until the change beats baseline, then ships via your target platform: a Langfuse prompt version, a GitHub PR, a Paperclip queue.
What it does
- Takes diagnostics results and known issues
- Generates and applies the mutation to your platform
- Validates against the evaluator, verifies post-update
Inputs
- Diagnostics Agent output
- Coding-agent runtime
- Target platform credentials
- Evaluator framework hook
Outputs
- Applied change: prompt, tool, code, or config
- Pre-merge validation via the evaluator framework
- Auto-application to your platform, audit trail with rollback
Works with
GitHub
GitLab
Langfuse
Claude
CursorGet access to Improve
Join our enablement program and we will set you up with the full Mutagent loop.