Benchmarks Skill (Internal Maintainer)

Use this for repository maintenance workflows. It is not a user-facing Graphistry domain skill.

Use This Skill For

• Running eval sweeps across journeys (skill pressure, persona, guardrails, etc.)
• Validating baseline isolation (skills=off must not read skill files)
• Generating public-safe benchmark reports
• Updating README.md with fresh benchmark numbers
• Creating CHANGELOG.md entries for benchmark releases
• Git version tagging (semver vX.Y.Z)

Success Criteria (Do Not Skip)

• Sweeps complete without baseline contamination (verify via log inspection)
• pass_bool is used for pass rate calculation (not score >= 0.8)
• Reports are public-safe (source paths redacted)
• README.md and benchmarks/README.md are updated with new numbers
• CHANGELOG.md has entry for the sweep
• Git tag created after merge