benchmark-triage

To analyze benchmark.json files, this skill reads iteration directories containing eval artifacts like grading.json and eval_metadata.json, synthesizes failure patterns, and outputs a structured Markdown report with recommended next edits.

Triage benchmark results by directing this skill to your iteration directory; it automatically identifies failure clusters and flaky patterns, then documents root causes with cited evidence from representative evaluative cases.

Generate a triage report by providing the path to your benchmark.json or iteration directory; the skill processes grading.json and eval_metadata.json artifacts to produce concise, actionable guidance for skill iteration.

Yes, you can get automated recommendations by running this skill on benchmark directories; it analyzes evaluation artifacts, highlights high variance or repeated failures, and outputs prioritized next edits in a Markdown report.

Benchmark triage works standalone without additional dependencies; it directly reads local benchmark.json and related evaluation artifacts from your iteration directories to synthesize patterns and generate structured reports.

Use benchmark triage reporting when you need to summarize benchmark results, identify representative deep dives into repeated failures, and translate chaotic evaluation artifacts into clear, prioritized guidance for improvement.