Codex Chef
Codex Chef is an unofficial community starter, not an OpenAI product. Its Codex guidance is based on official Codex documentation and is checked against the repository's current contracts. It supports Windows, macOS, Linux, and WSL, and keeps destructive, credentialed, database, publishing, deployment, and broad-filesystem actions behind explicit approval.
There are six README entry points for orientation. English and Turkish deep docs contain the complete operator guidance; the other language pages are concise, human-written routes into those canonical guides.
Start With A Preview
Check the local prerequisites first:
Get-Command git
Get-Command node
Get-Command npx
Get-Command codex
node -v
Node.js 18 or newer is required. If a command is missing, start with Troubleshooting.
Clone and preview without changing your Codex home:
git clone https://github.com/ucsahinn/codex-chef.git
cd codex-chef
npm run chef -- --install
Install only after the preview looks right:
npm run chef -- --install --apply
macOS, Linux, or WSL:
git clone https://github.com/ucsahinn/codex-chef.git
cd codex-chef
npm run chef -- --install
npm run chef -- --install --apply
The installer backs up managed targets before replacement. Normal install and repair do not prune user-owned skills, MCPs, profiles, or unrelated plugin files.
The colorful CLI is the recommended public entry point. Advanced/manual automation can call
scripts\install.ps1 on Windows or scripts/install.sh on Bash systems; the manifest-backed
operation contract is available through node scripts/plan-install.mjs --all --json --redact-paths.
Use The CLI Without Guessing
Run npm run chef for the numbered command center.
| Need | Command |
|---|---|
| Fast repository health | npm run chef -- --status --repo-only --no-log |
| Full status board | npm run codex:status |
| Install preview | npm run chef -- --preview |
| Routing map | npm run chef -- --routing --profile starter-health |
| Diagnostics | npm run chef -- --diagnostics --no-log |
| Process audit | npm run chef -- --processes |
| Repair managed files | npm run chef -- --repair, then npm run chef -- --repair --apply |
| Update preview | npm run chef -- --update --plain --no-log |
| Update and verify | npm run chef -- --update --apply |
Inspection commands stay read-only. Write-capable commands require --apply or an action-specific typed confirmation. Use --no-log when you do not want the CLI to create its normal redacted, repo-local diagnostic record.
Detailed CLI behavior lives in Installation, Expected output, and Troubleshooting.
What Lands On Your Machine
| Surface | What you get |
|---|---|
| Agents | 21 named specialist role files. They are available for bounded delegation, not always-running background services. |
| Skills | Six bundled plugin workflows, including the project-scoped Codex Chef Brain, and sixteen reviewed optional global skills. Skills load when the task matches; they do not execute by themselves. |
| MCPs | Eight useful defaults for official docs, current library docs, reasoning, browser evidence, semantic navigation, memory reads, and local codebase graph reads through codebase-memory. Eight account, database, production, and broad-filesystem connectors stay off. |
| Guidance | A durable ~/.codex/AGENTS.md, routing profiles, approval rules, and token-safe profile choices that do not pin every agent to one model. |
| Safety | Dry runs, manifest-backed install plans, backup-first replacement, secret scanning, package-surface checks, and runtime verification. |
Installed skills do not execute by themselves; Codex loads them when a request matches their description or the user names one. Agent role selection is automatic, but spawning is conditional: delegation is reserved for independent parallel work, noisy research isolation, or an explicit user request.
What It Deliberately Does Not Do
- It does not store secrets, import browser sessions, copy private memories, or send telemetry to a maintainer service.
- It does not enable GitHub, Figma, Linear, Notion, Sentry, Vercel, Supabase, or broad filesystem access by default.
- It does not commit, push, tag, release, deploy, publish packages, rotate credentials, or change GitHub settings.
- It does not delete user data to make validation pass.
See The Workflow
Trust Signals
npm run checkvalidates docs, installers, agents, MCPs, skills, routing, package contents, release metadata, supply-chain indicators, and security boundaries.- CI covers the Windows installer, Ubuntu with Node.js 18, and macOS with Node.js 24.
manifests/install-plan.jsondescribes the managed write surface before an installer runs.- Authenticated and high-risk connectors remain opt-in.
- The repository is
private: truein npm metadata, so this source-first project cannot be published to npm by accident.
Verify locally:
npm run check
npm run token:audit
npm run verify:skills:online
git diff --check
gitleaks detect --redact --no-banner --no-git --verbose
For long or repository-wide sessions, npm run token:audit shows which context surfaces carry the most weight. The optional token-safe.config.toml profile lowers verbosity and tool-output limits without disabling agents, skills, MCPs, memory, hooks, or apps. Agent roles do not pin a model; automatic model/reasoning selection follows the active user profile.
Documentation
- Documentation map
- Türkçe dokümantasyon haritası
- Installation
- Security model
- Skills and agents
- MCP catalog
- Public readiness
- Advisory sources
- Knowledge base
- Türkçe bilgi bankası
- Agent-readable index
Contributing And Support
Read CONTRIBUTING.md before opening a change and SUPPORT.md before sharing diagnostics. Security reports belong in the private route described by SECURITY.md.
MIT licensed. Community maintained.
No comments yet
Be the first to share your take.