flow-skill-fix-by-benchmarks

To identify the root cause of a failing skill, run the benchmark framework on the target AssistFlow skill. It systematically pinpoints failures through structured benchmarking, interprets evidence, and proposes argumentation for fixes.

Debugging a skill that regressed after changes involves locating relevant benchmarks and running the benchmark-runner. This structured verification isolates specific failure points and prescribes steps to propose corrective actions.

You cannot apply benchmark fixes automatically to underperforming skills. The framework enforces guardrails by requiring explicit user approval before implementing any proposed fixes or changes to the skill.

Yes, this benchmarking approach works for skills failing validation runs. It applies directly to AssistFlow skills that underperform in validation runs, using evidence from the benchmark-runner to guide structured root-cause analysis.

You need benchmark-based verification for skill debugging when an AssistFlow skill fails benchmarks, regresses after changes, or underperforms in validation runs, requiring structured evidence interpretation to propose fixes.