af-fix-by-benchmarks

Root cause analysis for failing skill benchmarks requires executing benchmark scenarios to reproduce failures and collecting evidence from the results to pinpoint the underlying issue.

Benchmark-driven verification for skill failures uses a structured workflow to execute scenarios, collect evidence, and propose fixes with argumentation and verification plans to ensure stability.

To fix failing skills using benchmark scenarios, execute the benchmark runner, collect evidence from the results, and present a proposed fix with argumentation and a verification plan for user approval.

You can automate skill failure diagnosis using a subagent workflow that executes benchmark scenarios, collects evidence, and generates verification plans to ensure stability.

Yes, the benchmark-driven workflow pauses for user approval before implementing changes, presenting findings with a proposed fix, argumentation, and a verification plan.