Summary

This skill guides agents to gather evidence and diagnose root causes before making code changes. It enforces a diagnostic-first workflow: observe symptoms, trace inputs and state, rank hypotheses by evidence, and only proceed with fixes when a credible mechanism is confirmed. The skill is purely instructional with no file writes, shell execution, or external requests.

Trigger Keywords

Phrases that agents use to match this skill to user intent.

Use Cases

• Diagnose intermittent failures without speculating on causes
• Investigate performance regressions by tracing evidence
• Root-cause analysis for unknown or ambiguous errors
• Prioritize debugging hypotheses before implementing fixes
• Gather evidence to support or falsify debugging assumptions

Quality Notes

• Skill is action-oriented with clear decision criteria: separate symptom from cause, trace key data flows, rank hypotheses, stop exploration when evidence is sufficient
• Concise and memorable instructions using imperative language and parallel structure
• Well-scoped to investigation and diagnosis only — explicitly prohibits editing until cause is confirmed
• No file writes, shell commands, or network operations — safe for exploration phase
• OpenAI agent configuration provided with appropriate default prompt
• Limitations are implicit: the skill is diagnostic guidance, not a technical tool, so it does not include error handling or edge cases (those are context-dependent)
• Could be strengthened by examples of evidence-gathering patterns (e.g., 'check logs for error messages', 'compare behavior before/after regression') and criteria for 'sufficient evidence'