An AI-assisted bug triage workflow that does not invent a diagnosis

01 / Scenario

Start with the boundary, not the model.

Imagine that patients can double-book a physical slot near the spring daylight-saving transition, but only on reschedule in Europe/Bucharest. A useful triage process preserves that narrow scope until evidence proves otherwise. It does not rewrite scheduling rules for every timezone because the symptom sounds temporal.

02 / Workflow

A repeatable sequence.

1. 01

   Make the report actionable

   Capture build, environment, user path, expected behavior, actual behavior, frequency, impact, and existing evidence. Separate what the reporter observed from what the team assumes.

2. 02

   Reproduce—or bound non-reproduction

   Record exact setup, inputs, timestamps, versions, and outputs. If it does not reproduce, return the matrix you tried and the missing signal needed next instead of declaring the issue invalid.

3. 03

   Compare the failing path with a control

   A nearby working path often narrows the search faster than reading the whole system. In the timezone example, new booking already normalizes through the scheduling authority while reschedule does not.

4. 04

   State the smallest supported cause

   Name the condition, the mechanism, and the evidence that connects them. List plausible alternatives that were ruled out and those not yet tested.

5. 05

   Design the minimal repair

   Correct the cause at the existing authority boundary. Preserve unrelated policy and avoid opportunistic cleanup unless it is required for safety.

6. 06

   Protect the invariant

   Write a regression that fails for the original defect, control tests for adjacent behavior, and rollout/rollback checks appropriate to the risk.

03 / Fictional worked excerpt

What a bounded output looks like.

The table below condenses a fictional example included for demonstration. It is not a customer result, production test, approval, or performance claim.

04 / Failure modes

Where otherwise plausible AI output goes wrong.

• Diagnosing from the ticket title
• Changing global restrictions to make one reproduction pass
• Writing a test that mirrors the implementation instead of the invariant
• Claiming production repair from a local or fictional example

05 / Read-only checklist

Use this before calling the artifact complete.

• Exact environment and build
• Expected and actual behavior
• Reproduction log or bounded non-reproduction
• Working control path
• Cause, mechanism, and supporting evidence
• Minimal repair boundary
• Regression plus control cases
• Rollout signal and recovery path

This checklist is intentionally read-only: it helps you inspect a workflow without exposing the full native skill files, portable prompts, templates, or the bundle-exclusive orchestration skill.