Bug-fix verification

Objective

Convert a bug report into a small, demonstrably improved behavior without mistaking green tests for universal correctness.

Steps

1. Frame. Record observed symptom, expected behavior, relevant inputs, and scope. Distinguish user report from directly reproduced evidence.
2. Inspect. Locate responsible code, relevant configuration, and nearby tests. Look for alternative explanations before editing.
3. Reproduce. Attempt the smallest repeatable reproduction in a safe environment. Capture exact failing command, assertion, or observable result.
4. Regression. When practical, add or adapt a test that fails for the reported bug before the fix. Avoid contrived tests that only encode the implementation.
5. Fix narrowly. Change the minimum necessary behavior. Preserve public contracts, validation, security checks, and unrelated behavior.
6. Verify. Run the reproduction or regression again, then relevant adjacent checks. Record outcomes and whether the pre-fix failure was confirmed.
7. Inspect the diff. Check that tests meaningfully cover the defect and that no incidental files or abstractions were introduced.
8. Report. State reproduction status, evidence before and after, remaining coverage gaps, and any unreproduced conditions.

Decision rules

• If reproduction is impossible due to missing data or environment, do not fabricate a failure. Document the blocker, use a reasoned alternative check if available, and qualify the result.
• If a new test is artificial or disproportionately costly, explain why and use the most credible available verification instead.
• For destructive, production, credential, or security-sensitive reproduction, stop or use a disposable/authorized environment.
• Do not silently fix a separate issue discovered while debugging.

Output

Provide symptom, reproduction, fix, tests/checks, and residual uncertainty. Distinguish not reproduced, failed before/passed after, and merely passed after.