Terms in this guide: Evaluation · Prompt · Skill
A debugging collection that turns a vague failure into a reproducible case, an evidence-backed cause, and a verified correction.
Debugging is a chain of evidence
A useful debugging workflow moves from symptom to reproduction, from reproduction to cause, and from cause to a focused fix with verification. Skipping a link creates confident explanations that may describe a different failure.
The collection contains reproduce-issue, root-cause-trace, and fix-verification skills. Use the existing change-handoff skill after the correction to communicate results.
Reproduce the issue
Capture the observed and expected behavior, exact input, environment, relevant state, and the smallest reliable steps. Reduce the case while preserving the failure. Record whether it is deterministic, intermittent, or dependent on timing or external services.
Trace the cause
Follow the failing value, request, event, or rendered output through the smallest relevant path. Test competing explanations with targeted observations. Identify the responsible layer and the evidence that rules out nearby causes.
Implement a focused fix
Change the layer that owns the failure and preserve unrelated behavior. Avoid broad exception handling, global style overrides, or silent fallbacks that hide the symptom without correcting the cause. Keep the patch reviewable.
Verify against the reproduction
Run the original reproduction and observe the corrected behavior. Add a meaningful neighboring case or regression test when it protects an important contract. Confirm that the test would fail against the old behavior if practical.
Use the pack responsibly
These skills organize a general debugging method. They do not replace domain-specific safety, security, privacy, or data-integrity review. Escalate when the failure affects high-stakes systems or requires specialist access.
Keep each stage falsifiable
The reproduction skill should produce a sequence another person can follow from a defined state. The root-cause skill should explain the first incorrect value or decision and rule out plausible alternatives. The verification skill should rerun the original trigger and an adjacent path. If any stage cannot produce its evidence, stop and record the missing condition instead of passing a guess forward.
This separation prevents a common failure: changing code as soon as an error line appears. The line that reports a failure can be far from the value or assumption that caused it. A smaller reproduction makes the trace cheaper, and a causal trace makes the final regression check more durable.
Evaluate the pack on known defects
Choose a small bug with a stable trigger and known cause. Hide the cause from the first run, then judge whether the workflow reduces the failure, tests competing explanations, and identifies the correct boundary. Introduce an adjacent state to see whether verification only checks the happy path.
Record false starts as evaluation evidence, but keep the public skill concise. The package review should state which tasks, environments, and dates were used. AboutCodex currently labels these downloads as structure checked until representative workflow evaluations are completed.
Progress stays in this browser. No account needed.
Updates and upgrade guidance
- Learning context added
Added prerequisites, a suggested practice estimate, expected output, and connected definitions. The article procedure was not independently re-evaluated in this change.
What to do: Use the opening checklist to prepare. No software update is required for this editorial change.
This log starts with the first recorded review; it is not a reconstruction of earlier revisions.