analyze

Analyze is a thin routing skill that delegates to the Debugger agent the moment you describe a problem. Instead of manually spinning up a debugger sub-agent, you describe the symptom and analyze dispatches it with the right context, model tier, and evidence-gathering protocol already configured.

Purpose

Analyze exists so you never have to remember the agent name or construct the delegation prompt yourself. It maps "analyze", "debug", and "investigate" triggers directly to a structured root-cause workflow that reads logs, traces call paths, isolates the regression boundary, and reports a precise diagnosis.

How to invoke

Natural language triggers: "analyze", "debug", "investigate", "what's wrong with", "why is X failing".

Explicit slash: $analyze

codex
> analyze why the auth middleware is returning 403 on valid tokens

codex
> debug the memory leak in the job processor

What it does

• Dispatches a Debugger agent at the appropriate model tier
• Reads logs, stack traces, and relevant source files to build an evidence base
• Traces call paths to isolate the regression boundary or root cause
• Distinguishes symptoms from causes and reports the precise failure point
• Suggests a concrete fix or the next investigative step

Outputs

• Root-cause report identifying the failure point and evidence trail
• Reproduction steps or a minimal reproducer when discoverable
• Recommended fix or next investigative action

Related

• $trace — causal trace for multi-step system flows
• $build-fix — targeted fix for CI reds and type errors
• $deep-dive — two-stage pipeline for complex, multi-system investigations