Agent Definition Review Claude Skill

Find your agent's weakest layer before it finds you.

An operator's review of a finished agent creation prompt, run as a Claude skill: six layers force-ranked, every permitted action audited, and a verdict on single agent versus team.

What you'll get

• Rank all six layers, model, context, tools, control, guardrails, and observability, strictly 1 to 6 with quoted evidence, so you can say our weakest layer is X because Y and the first thing we do is Z.
• Audit every action on the read, draft, write, commit ladder against the human relationship each one really has.
• Verify the never list: every never enforced rather than wished, and every guardrail graded for enforceability and coverage.
• Get the architecture verdict: one agent or a team, with the proposed structure when it should be a team.