agent-evaluation

name	agent-evaluation
description	ALWAYS use this when the request matches Agent Evaluation: Testing and benchmarking LLM agents including behavioral testing, capability assessment, reliability metrics, and production monitoring—where even top agents achieve less than 50% on real-world benchmarks

Agent Evaluation

Selective Reading Rule

Start with:

• references/senior-master-standard.md
• references/usage-routing.md
• references/quality-checklist.md

Then load only the inherited docs, scripts, assets, examples, or detailed instructions that match the user's actual task.

Progressive Disclosure

• Read references/full-skill-instructions.md only when the task needs the full original upstream guidance.
• Prefer bundled scripts for deterministic or repetitive operations when they exist.
• Reuse bundled assets instead of inventing substitutes when assets are present.
• Keep final answers aligned with this skill's frontmatter trigger, Senior Master standard, and Codex strict review gate.