OpenAI Codex CLI: from zero to autonomous PR machine.

1# 1. Install via npm
2npm install -g @openai/codex
3 
4# 2. Sign in (opens browser, uses your ChatGPT Plus)
5codex login
6 
7# 3. Verify
8codex --version
9 
10# 4. Pick the model — gpt-5-codex is the default, you usually want it
11codex config set model gpt-5-codex

1cd ~/projects/my-saas
2 
3# Trust this repo for Codex
4codex
5# → in the TUI, press / and run /trust
6 
7# Or set the auto-approval mode for read-only commands
8codex config set approvalMode "on-failure"

1# AGENTS.md — Codex briefing
2 
3## What this repo is
4A Next.js 16 SaaS for [your product]. App Router. TypeScript. Tailwind v4.
5 
6## Build + run
7- `npm run dev` — http://localhost:3000
8- `npm run lint` — must pass before commit
9- `npm run test` — Vitest, runs in <30s
10- `npm run build` — clean production build
11 
12## Conventions
13- Server Components default, `'use client'` only when needed
14- Imports use `@/*` alias for `./src/*`
15- Server-only files import from `@/lib/server/*`
16- Public env vars must be `NEXT_PUBLIC_*`, others are server-only
17 
18## Important: do NOT
19- Commit .env or .env.local
20- Run destructive git commands without asking
21- Add dependencies without flagging them
22- Touch supabase/migrations/ — those run in production
23 
24## Where to look first
25- API endpoints: src/app/api/**/*.ts
26- DB queries: src/lib/db/*.ts
27- UI components: src/components/

1Run before I push a PR:
2 
31. `npm run lint` — fix every warning
42. `npm run test` — all green
53. `npm run build` — no errors
64. `git diff main` — scan for debug logs, hard-coded URLs, secrets
75. Stage staged changes, write commit message in conventional-commits style
86. Push to current branch
97. Open PR with: title, what changed, why, how I tested
10 
11If any step fails, STOP and report.

1# Edit ~/.codex/config.toml
2[mcp.servers.supabase]
3command = "npx"
4args = ["@supabase/mcp-server"]
5env = { SUPABASE_ACCESS_TOKEN = "sbp_..." }
6 
7[mcp.servers.vercel]
8command = "npx"
9args = ["@vercel/mcp-server"]
10env = { VERCEL_TOKEN = "..." }
11 
12[mcp.servers.github]
13command = "npx"
14args = ["@github/mcp-server"]
15env = { GITHUB_TOKEN = "ghp_..." }

1codex run "Add rate limiting to /api/contact.
2 
3Constraints:
4- Use Upstash Ratelimit (already in deps)
5- 5 requests per IP per minute
6- Return 429 with Retry-After header
7- Add a test in __tests__/api/contact.test.ts
8 
9Done when:
10- The new test passes
11- npm run lint is clean
12- npm run build succeeds
13- A PR is open with a clear description"