CodexKit

Open-source operating kit for people using OpenAI Codex and ChatGPT to think, write, analyze, decide, automate routine work, and ship better output with more consistency.

CodexKit is a fresh project rebuilt around the surfaces that matter in Codex today: local Skills, high-signal playbooks, automation recipes, operational templates, department starter workspaces, repo guardrails, and MCP onboarding guidance. The pack covers 82 skills across 13 domains: engineering workflows, high-reasoning work, and low-reasoning office automation across project management, finance, legal, HR, operations, supply chain, strategy, analytics, marketing, data, customer success, IT & admin, training & development, and cross-functional work. It follows the official Codex Skills layout with SKILL.md, optional agents/openai.yaml, and standard .agents/skills discovery paths.

What is included

• skills/: 82 installable Codex Skills across 13 domains — engineering, high-reasoning business work, and low-reasoning office automation.
• playbooks/: copy-ready prompts for clarify, execute, review, decision, delegation, and release work.
• automations/: recurring task recipes for engineering plus weekly business, close, hiring, legal intake, operations, and marketing routines.
• mcp/: practical guidance for choosing and rolling out MCP servers without overloading the team.
• templates/: reusable templates including department templates (PM, finance, HR, legal, operations, marketing, cross-functional), the SKILL.md v2 template, and a deep research master prompt for skill authoring.
• workspaces/: starter workspace kits for PM, finance, HR, legal, ops, and marketing teams.
• scripts/: cross-platform skill installers, workspace quick-start scripts, and a pack validator.
• web/: a Next.js docs site for publishing the kit as a public open-source project.
• skill-finder.md: situation-based skill index — "I need to… → Use this skill".
• HUONG-DAN-NHANH.md: Vietnamese quick-start guide for non-technical office users.

Who it is for

• Individual operators, analysts, managers, and developers who want Codex to behave more like a disciplined specialist.
• Teams using ChatGPT plus Codex-style workflows that need repeatable prompts, review standards, and operational templates across knowledge work.
• Office-heavy teams that want Codex to handle status assembly, follow-up drafting, intake routing, KPI packaging, and other repeatable workflow chores.
• Maintainers who want a publishable starter repo instead of a private pile of prompts.

Quick start

Fastest path for non-coders

If you do not want to use Git commands or remember terminal steps, use the GitHub release package:

1. Open the Releases page for this repository.
2. Download codexkit-starter-pack-v0.9.0.zip.
3. Unzip it anywhere on your computer.
4. On Windows, double-click START-HERE-WINDOWS.cmd.
   • If Codex is not installed, the script will show installation instructions.
5. Restart Codex.
6. In Codex, type /skills to confirm all 82 skills appear.
7. Open skill-finder.md to browse skills by situation.
8. Optional: double-click CREATE-WORKSPACE-WINDOWS.cmd to create a starter workspace.

For macOS or Linux, download the same release package, unzip it, then run:

bash ./START-HERE.sh

1. Install CodexKit skills

Windows PowerShell:

.\scripts\install-skills.ps1

macOS / Linux:

bash ./scripts/install-skills.sh

By default, both scripts copy every folder from skills/ into $HOME/.agents/skills, which matches the user-scope Codex Skills location documented by OpenAI.

Windows double-click install is also available:

• START-HERE-WINDOWS.cmd: installs the skills into %USERPROFILE%\.agents\skills
• CREATE-WORKSPACE-WINDOWS.cmd: asks for a workspace starter and destination folder

Shell shortcuts are also available for extracted release packages:

• START-HERE.sh
• CREATE-WORKSPACE.sh

For repository-scoped discovery, install the pack into .agents/skills inside the repo:

Windows PowerShell:

.\scripts\install-skills.ps1 -Destination .\.agents\skills

macOS / Linux:

CODEXKIT_DESTINATION=./.agents/skills bash ./scripts/install-skills.sh

Codex scans .agents/skills from the current working directory up to the repository root, then also checks $HOME/.agents/skills. If an update does not appear immediately, restart Codex.

Codex can use skills through explicit invocation or implicit description matching. In CLI and IDE workflows, use /skills or type $ to mention a skill directly. CodexKit keeps codexkit-cloud-delegation and codexkit-automation-designer explicit-only to avoid accidental activation on sensitive workflows.

If the install completed but the skills do not appear:

1. Restart Codex.
2. Check that the skills were copied into %USERPROFILE%\.agents\skills on Windows or $HOME/.agents/skills on macOS/Linux.
3. In Codex, use /skills to verify discovery.

2. Validate the pack

node ./scripts/validate-pack.mjs

3. Run the documentation site

Use Node 20.9+ and npm 10+ for the docs app and local validation workflow.

npm --prefix web install
npm run dev

4. Update to the latest version

Git users:

bash ./scripts/update-codexkit.sh

Windows:

.\scripts\update-codexkit.ps1

Or double-click UPDATE-WINDOWS.cmd.

For macOS / Linux, run:

bash ./UPDATE.sh

The update script auto-detects your install method:

• Git clone: Runs git pull + re-installs all skills with --force.
• Zip download: Fetches the latest release from GitHub, extracts new skills, and overwrites the installed ones.

5. Start from a department workspace

Copy one folder from workspaces/ into your own repo or operating folder, then adapt the files to your context. Each starter workspace is opinionated on cadence, core artifacts, and the mix of high-reasoning versus routine automation work.

You can also scaffold from the command line:

.\scripts\quick-start.ps1 -List
.\scripts\quick-start.ps1 -Starter project-management-office -Destination .\acme-pmo

bash ./scripts/quick-start.sh --list
bash ./scripts/quick-start.sh --starter finance-performance-desk --destination ./acme-finance

For Windows users who prefer prompts instead of command arguments:

1. Double-click CREATE-WORKSPACE-WINDOWS.cmd.
2. Copy the starter name from the list.
3. Enter a destination folder such as .\my-pmo or C:\Work\FinanceDesk.

Recommended adoption path

1. Install the skills into $HOME/.agents/skills or copy selected skills into repo-local .agents/skills.
2. Pick one starter workspace that matches your function and adapt its files to your real cadence.
3. Start with one high-reasoning skill and one low-reasoning automation skill that match your most common workflow.
4. Tailor the automation recipes to your repo and operating cadence.
5. Add the department templates you will actually reuse, not every template in the pack.
6. Publish the docs site after replacing any placeholder organization metadata.

Folder map

CodexKit/
|-- automations/
|-- mcp/
|-- playbooks/
|-- scripts/
|-- skills/
|-- templates/
|-- workspaces/
|-- web/
|-- skill-finder.md
|-- HUONG-DAN-NHANH.md
|-- START-HERE-WINDOWS.cmd / START-HERE.sh
|-- UPDATE-WINDOWS.cmd / UPDATE.sh
`-- CREATE-WORKSPACE-WINDOWS.cmd / CREATE-WORKSPACE.sh

Design principles

• Codex-first, not assistant-agnostic.
• Small set of sharp assets over a giant pile of generic prompts.
• High-reasoning knowledge work deserves first-class skills, not just code prompts.
• Routine coordination work should be automated with low-noise, low-drama skills and templates.
• Review output must be risk-ranked and actionable.
• Automations must include guardrails, not just schedules.
• MCP adoption should be intentional, observable, and reversible.

Skill architecture

CodexKit uses a 5-Layer Skill Framework with a tiered structure:

Layers

Every skill addresses five concerns: Intent (clear purpose), Knowledge (domain expertise), Execution (step-by-step procedure), Verification (4C quality gates — Correctness, Completeness, Context-fit, Consequence), and Evolution (versioned changelog).

Tiers

| Tier | Structure | Current coverage | |------|-----------|------------------| | 1 | SKILL.md + agents/openai.yaml | All 82 skills | | 2 | + verification/ + examples/ | 21 high-value skills | | 3 | + templates/ + scripts/ | 5 technical skills |

Categories

| Category | Skills | |----------|--------| | knowledge | 17 | | data | 14 | | scaffolding | 14 | | runbook | 10 | | verification | 9 | | automation | 8 | | review | 6 | | infra | 4 |

Authoring new skills

Use templates/SKILL-TEMPLATE.md as the starting point. Optionally, use templates/deep-research-prompt.md with an external AI research tool (ChatGPT Deep Research, Gemini, Perplexity Pro) to generate domain knowledge — the 6-section output maps directly into the SKILL.md sections. See skills/codexkit-skill-template/ for a working reference implementation.

Full authoring guide: CONTRIBUTING.md.

Contributing

CodexKit is open-source and community-driven. There are several ways to help:

• Enrich existing skills: 61 skills still need Tier 2 content (verification/, examples/). Pick a skill you know well, create domain-specific checklists and examples, and open a PR.
• Add Tier 3 assets: Technical skills benefit from scripts/ and templates/. Build a reusable helper or output template.
• Create new skills: Use templates/SKILL-TEMPLATE.md + templates/deep-research-prompt.md to author a skill for a domain you are expert in.
• Add gotchas: If you encounter a real-world edge case while using a skill, document it using `templates/gotchas-