Repo workflow proof
Codex in this repo
Codex is the coding agent. This page shows the repo setup I actually use: guardrails, local Skills, verification, and human review.
This is not a documentation mirror. For current Codex product behavior, use the official docs. This route exists to show how the workflow is implemented and governed in this repository.
How this repo uses Codex
One bounded workflow, one proof surface, and one handoff to the official docs.
Start with a bounded task
The repo treats Codex as a coding agent for a concrete change, not a free-form chat demo.
Load repo guardrails
AGENTS.md defines scope, approval triggers, and the working rules before edits begin.
Attach reusable setup
Repeatable workflows live in repo-local Skills instead of scattered one-off prompts.
Patch and verify
Changes are expected to run through build, test, and Codex-specific checks.
Keep HITL visible
Human approval stays explicit for consequential changes and proof-sensitive routes.
Implementation proof
Everything below exists in this repo today.
AGENTS.md
This repo uses persistent project guidance for scope, review boundaries, and verification expectations.
.agents/skills/
Repo-local Skills handle repeatable work such as project pages, SEO, accessibility, and Codex context.
Verification
Patch work runs real checks including verify:patch, test:ci, check:build, check:codex, and codex:skills.
Route proof
This page is here to show implementation judgment in this repo, not to restate product docs that already live on OpenAI.
Why it matters
- Persistent guardrails reduce prompt drift.
- Repo-local Skills make repeatable workflows easier to reuse and review.
- Verification plus visible human approval makes the workflow easier to trust.