AI‑Powered Defect Analyzer: A Structured Full‑Chain Bug Investigation Skill

Why a Structured Approach Is Needed

A production incident cost ¥30,000 because a tester assumed a slow database query on the coupon table was the root cause. After adding an index and optimizing SQL, the problem persisted. The real issue was a mis‑configured circuit‑breaker: the gateway cut the connection at 3 seconds while the coupon service timed out after 5 seconds, leaving the request hanging.

The core enemy of defect analysis is confirmation bias – seeing a clue, forming a story, and then only seeking evidence that supports that story.

The defect-analyzer Skill

This skill splits defect analysis into eleven mandatory parts, forcing a disciplined workflow.

1. Defect Phenomenon

Operation: User selects a coupon on the order page and clicks “Submit Order”.

Expectation: Order completes within 2 seconds and redirects to the payment page.

Actual: UI spins for >10 seconds, shows “System Busy”, and no order is created.

2. Impact Scope

Severity: P0 – core ordering function broken.

Frequency: High‑frequency intermittent; started at 10:00 am, multiple users reported, not 100 % reproducible.

3. Known Facts (Evidence)

Page spins >10 seconds – user screenshots & customer‑service feedback.

Final error is “System Busy” – user screenshots.

Orders succeed without coupons – partial user feedback.

Problem started at 10:00 am – timeline from support tickets.

4. Unknown Information

Questions that still lack answers, e.g., exact reproduction steps, error‑code details, affected versions, user‑segment patterns.

5. Possible Causes (Likelihood & Evidence)

Coupon‑deduction DB slow query / deadlock (High) – no‑coupon orders work, 10‑second spin suggests lock wait.

Coupon service timeout without degradation (High) – “System Busy” indicates timeout but circuit‑breaker didn’t fire.

Coupon status‑check bug (Medium) – order not generated, possible infinite loop in validation.

Database connection‑pool exhaustion (Medium) – sudden spike at 10:00 could be batch jobs consuming connections.

Frontend parameter error (Low) – no direct evidence, only theoretical.

6. Verification Methods

Slow query / deadlock: Search slow‑query logs after 10:00 for coupon SQL. Expected result: execution time >10 s or lock‑wait entries. Cost: Low.

Service timeout without degradation: Inspect the call‑chain of “Submit Order” for coupon‑service latency and circuit‑breaker status. Expected result: call exceeds timeout and breaker not triggered. Cost: Medium.

Status‑check bug: Reproduce in a test environment with a debugger, watch checkCouponStatus(). Expected result: thread hangs inside the method. Cost: Medium.

7. Recommended Investigation Order

Check slow‑query logs & connection‑pool metrics (≈5 min) – validates causes 1 and 4.

Inspect call‑chain monitoring for timeout & breaker (≈10 min) – validates cause 2.

Add debug logs in test env to reproduce and observe behavior (≈30 min) – validates causes 1‑3.

Ask ops for change records around 10:00 am (meeting required).

The rule is “do the cheap things first”: high‑likelihood, low‑cost items are investigated before expensive or low‑likelihood ones.

8. Additional Information Needed

Error logs for the “Submit Order” API after 10:00 am (ELK export).

APM P50/P99 response‑time distribution.

Full trace IDs for failed requests (trace platform).

Slow‑query statements >1 s after 10:00 am (DBA).

Code/configuration changes in the 24 h before 10:00 am (release platform).

9. Possible Fix Directions

Option A – Optimize slow query: Add index / reduce joins (root‑cause fix, may affect other features).

Option B – Adjust circuit‑breaker: Set timeout to 2 s and degrade gracefully (quick stop‑bleed, requires compensation for coupon validation).

Option C – Temporarily disable coupons: Deploy config to hide coupon entry (5 min rollout, impacts user experience).

10. Post‑Fix Validation

Functional verification: normal flow, error flow, boundary values.

Regression: full coupon test suite + order‑flow suite.

Data check: ensure no “order created but coupon not deducted” dirty data.

Monitoring: P99 response time < 2 s.

Long‑term observation: watch 24 h for new “System Busy” incidents.

11. Regression Test Points to Prevent Recurrence

Concurrent coupon usage – only one of two orders should succeed (P0).

Coupon‑service timeout – verify degradation works (P0).

Boundary: coupon value equals order amount.

Performance: coupon‑deduction P99 < 500 ms (P1).

Quality Red Lines (Hard Constraints)

Never conclude without evidence.

Never mix fact and hypothesis.

Always list ≥2‑3 possible causes.

Provide a verification method for every hypothesis.

Always specify impact severity.

How to Use the Skill

Place the defect-analyzer-cross-platform folder under ~/.workbuddy/skills/ and invoke it with a prompt like “Analyze this defect”. Python helper scripts are also supplied to generate the 11‑part framework from a markdown description or to check an existing draft for missing sections.

Conclusion

By forcing analysts to record facts, enumerate hypotheses, rank by likelihood × cost, and verify each step, the defect-analyzer skill turns intuition‑driven debugging into a repeatable, bias‑resistant process, saving time and preventing costly mis‑diagnoses.