Fix model regressions in your agent skills.
What it does
Retunes an agent instruction corpus for a target model so its measured behavior meets a pre-registered bar.
When to use it
Use when an instruction corpus degrades on a new model.
How to use it
Give it a target model or regression symptom and an instruction corpus; it produces corpus changes measured against a pre-registered bar.
What you provide
- target model or symptom
- An existing project
Good to know
- starts one helper agent per item, so the count depends on the work
- cannot be undone: benchmark harness
Latest review
What you need · 1
Requires a benchmark harness that can compare two corpus builds, access or produce detailed run logs, and run a repeatable end-to-end task.
Access · 2
This skill
benchmark harness
Execute
Paid model evaluations (irreversible)
pull requests
Read
Reads pull requests for provenance