Each phase should answer a different question: what exists, what will change, did it change, and does it work?
The decision behind the technique
Skipping exploration creates brittle assumptions; skipping verification leaves a plausible diff with unknown behavior. The loop keeps scope and evidence visible.
Each phase should answer a different question: what exists, what will change, did it change, and does it work?
A practical step-by-step workflow
Use this sequence as a default, then tighten or expand it only when the evidence changes.
- Explore architecture and current behavior without editing
- Plan the smallest coherent change and checks
- Implement inside the agreed boundary
- Run checks, exercise the path and review the diff
Failure boundary and finish line
If verification contradicts the plan, return to evidence instead of stacking speculative patches.
Finish when the next action, owner, evidence and stop condition are all explicit.
Frequently asked questions
What should I do first?+
Explore architecture and current behavior without editing
What is the main failure mode?+
If verification contradicts the plan, return to evidence instead of stacking speculative patches.
OFFICIAL SOURCES
Verify against the source
OpenAI Cookbook: development workflows with CodexOpenAI Codex best practicesThis guide provides general information, not an OpenAI allowance promise or support decision. Your account and official pages govern actual usage, benefits and recovery times.