Codex Chef
This is an unofficial community starter, not an OpenAI product. It is mapped to current official Codex documentation and keeps destructive, credentialed, publishing, deployment, database, and broad filesystem actions approval-gated.
The multilingual README entry points, six-language deep docs coverage, and bilingual knowledge base are part of the release surface. English and Turkish keep the most complete operator flow; Deutsch, Español, Português (Brasil), and Français provide concise local entry points and links into the same deep docs.
Copy-Paste Install
Before you paste, confirm the local prerequisites so setup failures point to the right layer:
Get-Command git
Get-Command node
Get-Command npx
Get-Command codex
node -v
Node.js 18 or newer is required. If one of these commands is missing, use Troubleshooting before treating the repository as broken.
Preview first:
git clone https://github.com/ucsahinn/codex-chef.git
cd codex-chef
powershell.exe -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1 -All -WhatIf
node scripts/plan-install.mjs --all --json --redact-paths
Install after reviewing the preview:
powershell.exe -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1 -All -Interactive
Bash or WSL:
git clone https://github.com/ucsahinn/codex-chef.git
cd codex-chef
chmod +x scripts/install.sh
./scripts/install.sh --all --interactive
Repair an existing Codex Chef setup without deleting user skills:
powershell.exe -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1 -Repair -WhatIf
powershell.exe -NoProfile -ExecutionPolicy Bypass -File .\scripts\install.ps1 -Repair
CLI Without The Noise
Use npm run chef for the menu. Keep these commands handy:
| Need | Command |
|---|---|
| Fast local health check | npm run chef -- --status --repo-only --no-log |
| Full status board | npm run codex:status |
| Install preview | npm run chef -- --preview |
| Routing profile map | npm run chef -- --routing --profile starter-health |
| Diagnostics hub | npm run chef -- --diagnostics --no-log |
| Process audit only | npm run chef -- --processes --no-log |
| Repair preview/apply | npm run chef -- --repair then npm run chef -- --repair --apply |
Rules of thumb:
--status,--preview,--skills,--mcp,--routing,--diagnostics,--processes,--auth, and--logsare inspection-first surfaces.- Write paths require
--applyor typed confirmation. The menu labels the write boundary before doing anything risky. - Add
--no-logwhen you want a strict no-log audit. Normal CLI logs are repo-local, ignored, and redacted. - Do not use
npx run; usenpm run chefornpm --prefix <repo> run chef.
Detailed CLI behavior lives in Install, Expected output, and Troubleshooting.
What You Get
Codex Chef installs a reviewed Codex baseline, not a hidden copy of someone else's machine.
| Surface | What lands on your machine |
|---|---|
| Agent team | 21 Codex subagent role files under ~/.codex/agents/*.toml, including readable nickname_candidates. They are role definitions, not always-on services. |
| Durable guidance | Global ~/.codex/AGENTS.md with routing, verification, safety, and approval rules. |
| MCP defaults | 8 useful MCP entries enabled for docs, code navigation, browser evidence, reasoning, non-secret memory, and local codebase graph reads; mutating tools are prompt-gated or disabled, and 8 account/database/high-risk connectors stay disabled. |
| Plugin + skills | Local codex-chef-workflows plugin, four bundled skills, and sixteen reviewed optional global skills. |
| Safety gates | Dry runs, backups, validation, secret scanning, and approval gates before risky actions. |
Installed skills do not execute by themselves. They enter context when named by
the user or when a task clearly matches their description. Agent roles are
selected automatically, but spawning is conditional: use a subagent only for
genuinely independent parallel work, to isolate noisy logs or research from the
main thread, or when the user explicitly requests delegation. max_threads = 10 remains a capacity ceiling for multi-window use; ordinary routing targets
one to four focused agents. Agent files do not pin a model or reasoning effort,
so token-safe.config.toml and the user's active profile model/reasoning choices
remain authoritative.
Installed Agent Team
These are the visible specialist names Codex Chef installs. They are role files for explicit, visible delegation; they are not always-on background services.
| Map and plan | Build and review | Verify and ship |
|---|---|---|
Code Mapper (code_mapper) |
Design Reviewer (design_reviewer) |
QA Lead (qa_lead) |
Docs Researcher (docs_researcher) |
DevEx Auditor (devex_auditor) |
Performance Auditor (performance_auditor) |
Context Architect (context_architect) |
Root-Cause Debugger (root_cause_debugger) |
Google SEO Auditor (google_seo_auditor) |
Prompt Architect (prompt_architect) |
Docs Author (docs_author) |
Security Auditor (security_auditor) |
MCP Integrator (mcp_integrator) |
Spec Author (spec_author) |
Test Verifier (test_verifier) |
Product Strategist (product_strategist) |
Code Reviewer (code_reviewer) |
Release Verifier (release_verifier) |
Engineering Planner (engineering_planner) |
Frontend Verifier (frontend_verifier) |
Codex Doctor (codex_doctor) |
Skills Included
| Set | Skills |
|---|---|
| Local plugin | codex-chef-operator, offline-diagram-triplet, context-budget-planner, adaptive-agent-routing |
| Reviewed global catalog | dependency-upgrade, gh-fix-ci, systematic-debugging, request-refactor-plan, security-best-practices, frontend-skill, webapp-testing, web-quality-audit, seo, accessibility, test-driven-development, documentation-and-adrs, mcp-builder, ai-project-starter, prompt-architect, ai-skill-create |
MCP Defaults
| Status | MCPs | Boundary |
|---|---|---|
| Default-enabled narrowed tools | OpenAI Docs · Context7 · Sequential Thinking · Playwright · Chrome DevTools · Serena · Memory · codebase-memory |
Docs and reasoning run smoothly; browser evidence, semantic navigation, local memory reads, and local graph reads are allowlisted, while interaction, symbol edits, graph indexing, and writes stay prompt-gated or disabled. |
| Disabled until opt-in | Filesystem · GitHub · Figma · Linear · Notion · Sentry · ▲ Vercel · Supabase | These can expose private files, accounts, deployments, telemetry, or databases. |
Run npm run codex:status after install to see MCP setup notes, effective controls, routing profiles, and installed-runtime drift without mutating global Codex state.
What It Does Not Do
- It does not store secrets, copy private memory state, or import existing sessions; Memory MCP is only for non-secret local context when you use it deliberately.
- It does not enable authenticated account, database, production, or broad filesystem MCP connectors by default.
- It does not commit, push, tag, release, publish packages, deploy, rotate secrets, or change GitHub settings.
- It does not delete user data as a cleanup shortcut; repair and force flows are previewed and backup-backed.
Visual Overview
Trust Signals
| Signal | Evidence |
|---|---|
| Real validation | npm run check runs repo, docs, install-plan, installer smoke, agent drift, MCP drift, token-surface, skill-source, supply-chain, package-surface, release-readiness, and security checks. |
| Multilingual surface | Six root READMEs plus six-language deep docs are validated. |
| Operator KB | kb/ keeps short English and Turkish task articles for install preview, runtime verification, routing, and release hygiene. |
| Conservative connectors | Account, database, production, and broad filesystem connectors stay disabled until a task explicitly needs them. |
| Install plan | manifests/install-plan.json and schemas/install-plan.schema.json define the managed write surface. |
| Context budget | npm run token:audit separates always-loaded instructions, discoverability metadata, invoked/deferred surfaces, repository maintenance size, tool schema/context, measured session telemetry when available, and per-agent cost. It is a context estimate, not provider billing. |
| Agent-readable index | llms.txt gives agents a compact map of install targets, docs, safety boundaries, and comparison sources. |
Repository Layout
.github/ CI workflow plus issue and PR templates
assets/ Public-safe README visuals
catalog/ Agent, skill, MCP, and routing metadata
README*.md Multilingual public entry points
docs/ Six-language setup and verification guides
kb/ English and Turkish task-focused knowledge base
manifests/ No-write install plan metadata
plugins/ Bundled local Codex plugin and skills
schemas/ Lightweight validation schemas
scripts/ Install, doctor, status, CLI, and validation scripts
templates/codex/ Files copied into ~/.codex
templates/git/ Optional global Git hygiene files
Verify Locally
npm run check
git diff --check
gitleaks detect --redact --no-banner --no-git --verbose
Optional online skill-source check:
npm run verify:skills:online
Documentation
Use the README for orientation; use the docs for operator detail:
- Install
- How to run the setup
- Expected output
- Troubleshooting
- Upgrade guide
- Codex capability map
- Workflow surface map
- Codex surfaces
- Codex flags
- Skills and agents
- MCP catalog
- Security model
- Verification
- Public readiness
- ECC compatibility
- SEO and discoverability
- Research notes
- Advisory sources
- Publishing
- Knowledge base
- Turkish knowledge base
- Agent-readable index
Official Codex References
- Codex manual: https://developers.openai.com/codex/codex-manual.md
- Skills: https://developers.openai.com/codex/skills
- Plugins: https://developers.openai.com/codex/plugins
- MCP and connectors: https://developers.openai.com/api/docs/guides/tools-connectors-mcp
- Windows: https://developers.openai.com/codex/windows
- Config, permissions, rules, hooks, and
AGENTS.mdmapping: Research notes
Publishing Boundary
This repo is built to be public-ready after validation, but the installer is local-only. Commit, push, tag, release, package publishing, deployments, and GitHub settings changes must be explicit human decisions after local verification.
No comments yet
Be the first to share your take.