A practical field guide · English
Make the idea clearer. Make the work real.
A calm, evidence-led way to work with Codex and other AI: describe the destination, share the right context, protect what matters, and prove the result.
Choose a doorway
Use the page that matches the work in front of you.
You do not need to read this like a textbook. Follow the shortest path to the decision, prompt, skill, or quality check you need.
Working with Codex
What Codex can do, where it gets context, and how to keep yourself in control.
PromptsAsk for outcomes
Reusable templates for ideas, research, debugging, coding, review, and long-running work.
BuildReach the result
Turn a vague idea into a bounded project with acceptance criteria and a verification loop.
SkillsChoose the right skill
Find skills, decide when they apply, and avoid loading a large workflow when a small one is enough.
TrustReview a skill
A practical rubric for triggers, scope, quality, safety, evidence, and real behavior.
ProofKnow when it is done
Match the evidence to the claim: tests, browser behavior, artifacts, sources, and final diff.
A guide shaped around your working style
The interaction advice is based on patterns visible in our recent work: you prefer hands-on progress after a short evidence pass, bounded changes over vague recommendations, plain explanations, careful preservation of existing systems, and a precise report of what was actually checked.
One useful rule
Do not optimize for impressive output. Optimize for a result you can inspect.
An answer can sound confident and still be wrong. A code change can look elegant and still break a permission boundary. A beautiful image can still miss the requested style. The guide therefore keeps returning to three questions:
Can another person recognize success without guessing?
Which behavior, privacy rule, design choice, or user preference is non-negotiable?
Choose it before the work is declared complete.