Debugging requirements protocol from reproducible issue to verified fix and regression protection.
<when_to_use>
Investigating functional bugs, crashes, regressions, or production incidents
Need structured debug flow for coding agent execution
Need explicit completion criteria for bug fixes
</when_to_use>
<input_requirements>
Symptom statement and impact
Expected vs actual behavior
Minimal reproducible example (steps/data)
Environment fingerprint (version, config, flags)
Evidence pack (errors, logs, traces, timestamps)
</input_requirements>
<debug_workflow>
Triage impact and confirm current reproducibility
Build minimal reproducible case
Localize failing layer and narrow search space
State testable hypothesis and instrumentation plan
Apply minimal root-cause fix
Verify fix against acceptance criteria
Add regression test/eval that fails before and passes after
Record residual risks and follow-up actions
</debug_workflow>
<quality_rules>
No reproducible signal means no claim of confirmed fix
Bugfix is incomplete without regression protection
Fix must target root cause, not only symptom masking
Verification must include evidence from commands or telemetry
</quality_rules>
<required_signals>
Repro rate before and after fix
Error rate or failure count change
Latency/perf guardrail impact when relevant
Regression suite result for affected path
</required_signals>
<do_not>
Do not debug by guesswork without hypothesis
Do not close issue without reproducible verification
Do not skip regression test because fix "looks obvious"
Do not merge broad refactor as hidden hotfix
</do_not>
<output_requirements>
Problem statement with expected vs actual
Root cause summary and impacted components
Fix summary with evidence of verification
Regression artifact added and location
</output_requirements>
Stack Overflow MRE Guide
MIT Debugging Workflow
OpenAI Evaluation Best Practices
OpenAI Evals Guide
Anthropic: Demystifying Evals for AI Agents
Google SRE Postmortem Culture
OpenTelemetry Context Propagation
