agentkaizen

Use agentkaizen to measure and prove whether your AI coding agent actually follows instructions. Works natively — no installation required. Use this skill when: you want to score a recent Codex or Claude Code session and check rule compliance (did it branch before work? stay within tool limits?); you changed AGENTS.md or config and need before/after evidence; a session used too many tool calls or failed to complete; you want a qualitative blind comparison of two agent outputs. Trigger: any question about measuring, comparing, or verifying agent behavior.