Skip to content
Current page: AI-Assisted Bug Fix Workflow with Review

Fix a reproducible bug with an AI coding agent

Capture a reliable reproduction, let a coding agent implement a focused fix, and verify the regression evidence before merge.

A good agent-assisted bug fix starts with evidence: the failure, expected behavior, relevant constraints, and the exact proof that would show the root cause is gone.

Prerequisites

Connect the affected repository and write a card with the shortest reliable reproduction, logs or a failing test, expected behavior, edge cases, and a bounded definition of done. If the root cause is not yet clear, add a read-only Research step before implementation.

Suggested sequence

Use the focused-fix starter recipe for a Codex implementation followed by Reviewer, or build a recipe that adds QA for behavior-heavy regressions.

  • Reproduce and identify the root cause
  • Implement the smallest durable change
  • Add focused regression coverage
  • Review the patch against the original evidence

Artifacts and controls

The team reviews the branch, diff, tests, preview when useful, and pull request. Unrelated refactors, unverified assumptions, and missing coverage remain reasons to request changes rather than merge.

Bring every coding agent onto one shared plane.

Plan with your team, run Claude and Codex in isolated sandboxes, and review the work before it ships.

Start a project