By name: debug-reliability description: Diagnose production or development failures, rank root-cause hypotheses, and deliver minimal-risk fixes. Use when systems are broken, unstable, flaky, or showing regressions.
Debug Reliability
Restore stability quickly using evidence-driven debugging.
Required Inputs
AGENTS.mdPROJECT_CONTEXT.md- Symptoms, logs, traces, and recent changes
Workflow
- Reproduce the issue or establish a reliable observation boundary.
- Gather evidence from logs, metrics, traces, and diffs.
- Produce ranked hypotheses and disqualifying checks.
- Validate top hypothesis with minimal experiments.
- Apply smallest effective fix and verify recovery.
- Capture prevention tasks for Learning Engineer.
Reliability Quality Gates
- Root cause is evidence-backed, not speculative.
- Fix surface area is minimal and reversible.
- Regression checks are executed for adjacent risk.
Required Output
- Ranked hypothesis table.
- Root cause statement with evidence.
- Patch summary and verification results.
Handoff Contract
- QA: regression focus and stress scenarios.
- Learning: cause-action-result entry with guardrail proposal.
Constraints
- Do not broad-refactor during active incident response.
- Avoid introducing net-new architecture in emergency fixes.
- Make uncertainty explicit.
References
references/playbook.mdreferences/agent-source.mdreferences/agent-source.md