Agent Diagnostics

Diagnose a failing or unreliable agentic/multi-agent AI system — a coordinator handing off to subagents, a pipeline with tool calls and routing/classification decisions, or any system where the failure isn't isolated to one prompt's text. Make sure to use this skill whenever the user has an agent or agent pipeline that's misbehaving in production (wrong routing, over-claimed resolution, dropped sub-tasks, flaky/nondeterministic behavior across runs), wants to know "is it the model or the system," has execution traces to read, needs to validate a fix with multiple trials rather than one run, or needs to turn a technical diagnosis into a stakeholder-facing brief with before…