Skip to content
/

OpenAI Skills (Official)

openai-skills · openai/skills · ★ 20k · last commit 2026-05-21

Primitive shape 45 total
Skills 45
00

Summary

OpenAI Skills (Official) — Summary

OpenAI's official skills repository for Codex is the canonical collection of skills for the Codex coding agent, organized into three tiers (.system, .curated, .experimental) with a bootstrapping install skill ($skill-installer) and a broad library of deployment, integration, and developer-workflow skills.

Problem it solves: Codex agents need reusable, discoverable capabilities for common development tasks — deploying to Vercel/Netlify/Cloudflare, integrating with Figma/Linear/Notion/Sentry, writing security threat models, and creating new plugins — that would otherwise require manual setup each session.

Distinctive trait: The only official skills pack that ships a skill-installer system skill (auto-installed in Codex) that enables agents to list and install curated or experimental skills by name from within a running Codex session, plus the plugin-creator system skill for scaffolding new Codex plugin directories with proper structure.

Target audience: Codex (OpenAI's coding agent) users who want ready-made capabilities for deployment pipelines, third-party integrations, and developer workflows; also compatible via the Agent Skills standard format.

Scope: 5 system skills (always pre-installed), 40+ curated skills, and experimental skills; all organized in .system/.curated/.experimental tiers under skills/.

Differs from seeds: Unlike superpowers or agent-skills (workflow enforcement), openai/skills is a curated integration library — skills for interfacing with external services (Figma, Vercel, GitHub, Sentry, Linear, Notion) rather than a software development methodology. The tiered install model (system→curated→experimental) parallels a package manager rather than a session orchestrator.

01

Overview

OpenAI Skills (Official) — Overview

Origin

Official OpenAI repository for Codex skills. Stars: 20,338 with 30 contributors. License: per-skill (each skill directory contains its own LICENSE.txt). Last pushed 2026-05-21.

Purpose

From the README:

"Agent Skills are folders of instructions, scripts, and resources that AI agents can discover and use to perform at specific tasks. Write once, use everywhere." "Codex uses skills to help package capabilities that teams and individuals can use to complete specific tasks in a repeatable way."

The Agent Skills Standard

Explicitly references agentskills.io as the open standard behind the format. Skills in this repo are OpenAI's implementation of that standard for Codex.

Tier System

Three tiers of skills:

  1. .system — Always pre-installed in Codex (5 skills). Not user-installable; auto-installed on Codex launch.
  2. .curated — Community-contributed, reviewed, and endorsed by OpenAI (~40 skills). Install by name.
  3. .experimental — Newer, less-vetted skills. Install by folder path or GitHub URL.

Bootstrap Philosophy

The $skill-installer system skill enables a key pattern: Codex can discover and install new skills from within a running session by asking an agent that has skill-installer to list available skills or install a named one. This makes the skills collection self-expanding.

License

Per-skill LICENSE.txt files inside each skill directory. The README notes the per-skill license approach, meaning commercial and non-commercial skills can coexist.

Contributing

Community values: "Be kind and inclusive." Security contact: security@openai.com.

02

Architecture

OpenAI Skills (Official) — Architecture

Distribution

  • Type: Codex skills (Agent Skills standard format)
  • System skills auto-installed in Codex on launch
  • Curated skills install:
    $skill-installer gh-address-comments
    
  • Experimental skills:
    $skill-installer install the create-plan skill from the .experimental folder
    $skill-installer install https://github.com/openai/skills/tree/main/skills/.experimental/create-plan
    
  • After install: Restart Codex to load new skills

Required Runtime

  • gh CLI (for skills that use GitHub API — e.g., gh-address-comments)
  • Python 3 (for skills with Python scripts)
  • Various deployment CLIs depending on skill (Vercel CLI, Netlify CLI, Cloudflare Wrangler, etc.)

Repository Structure

openai/skills/
├── skills/
│   ├── .system/                    # 5 auto-installed skills
│   │   ├── imagegen/
│   │   ├── openai-docs/
│   │   ├── plugin-creator/
│   │   ├── skill-creator/
│   │   └── skill-installer/
│   ├── .curated/                   # ~40 community-reviewed skills
│   │   ├── aspnet-core/
│   │   ├── chatgpt-apps/
│   │   ├── cli-creator/
│   │   ├── cloudflare-deploy/
│   │   ├── define-goal/
│   │   ├── figma-*/                # 7 Figma skills
│   │   ├── gh-address-comments/
│   │   ├── gh-fix-ci/
│   │   ├── hatch-pet/
│   │   ├── jupyter-notebook/
│   │   ├── linear/
│   │   ├── migrate-to-codex/
│   │   ├── netlify-deploy/
│   │   ├── notion-*/               # 4 Notion skills
│   │   ├── openai-docs/
│   │   ├── pdf/
│   │   ├── playwright-interactive/
│   │   ├── playwright/
│   │   ├── render-deploy/
│   │   ├── screenshot/
│   │   ├── security-*/             # 3 security skills
│   │   ├── sentry/
│   │   ├── speech/
│   │   ├── transcribe/
│   │   ├── vercel-deploy/
│   │   ├── winui-app/
│   │   └── yeet/
│   └── .experimental/             # Newer, less-vetted skills
│       └── create-plan/

Skill Format

Agent Skills standard: folder + SKILL.md with YAML frontmatter + markdown instructions. Each skill may also contain:

  • scripts/ — Python/shell helper scripts
  • LICENSE.txt — per-skill license

Config Files

None at the repository level — no hooks.json, no plugin.json. Skills are loaded by Codex directly from $CODEX_HOME/skills/.

03

Components

OpenAI Skills (Official) — Components

Commands

None — no slash commands. Skills are triggered by description matching or explicit agent invocation.

Skills (45+ total)

System Skills (5 — always installed)

Name Purpose
imagegen Generate images via OpenAI image generation APIs
openai-docs Authoritative OpenAI developer documentation lookup via MCP/WebFetch
plugin-creator Scaffold new Codex plugin directories with proper structure
skill-creator Create and iterate on new skills with eval loops
skill-installer List and install curated/experimental skills from within Codex

Curated Skills (~40, selected representative set)

Name Purpose
gh-address-comments Address GitHub PR review comments via gh CLI
gh-fix-ci Diagnose and fix failing CI runs
figma-use Use Figma designs in code
figma-generate-design Generate Figma designs from code/description
figma-implement-design Implement a Figma design as code
figma-create-design-system-rules Create design system rules from Figma
figma-code-connect-components Connect Figma components to code
figma-create-new-file Create new Figma files
figma-generate-library Generate Figma component library
vercel-deploy Deploy to Vercel
netlify-deploy Deploy to Netlify
cloudflare-deploy Deploy to Cloudflare Workers/Pages
render-deploy Deploy to Render
notion-knowledge-capture Save information to Notion
notion-meeting-intelligence Extract and save meeting notes to Notion
notion-research-documentation Document research findings in Notion
notion-spec-to-implementation Turn Notion specs into implementation
linear Manage Linear issues and projects
sentry Query and analyze Sentry errors
security-best-practices Apply security best practices to code
security-ownership-map Map code ownership for security review
security-threat-model Generate security threat models
playwright Browser automation with Playwright
playwright-interactive Interactive browser automation
screenshot Capture screenshots
jupyter-notebook Work with Jupyter notebooks
pdf Process PDF files
speech Text-to-speech and speech processing
transcribe Audio transcription
aspnet-core ASP.NET Core development patterns
chatgpt-apps Build ChatGPT apps with Apps SDK
cli-creator Create CLI tools
define-goal Define clear goals before starting work
migrate-to-codex Migrate projects to Codex
winui-app Windows UI app development
yeet Unknown/experimental
hatch-pet Unknown/experimental

Experimental Skills (1+)

Name Purpose
create-plan Create structured implementation plans

Subagents

None formally defined.

Hooks

None — Codex does not use the same hook system as Claude Code.

MCP Servers

The openai-docs system skill references the openaiDeveloperDocs MCP server (developers.openai.com/mcp) and includes an install command if missing. This is an external MCP server reference, not a bundled server.

Scripts

Multiple skills contain scripts/ directories with Python helper scripts:

  • skill-installer/scripts/list-skills.py — Lists available skills with install status
  • skill-installer/scripts/install-skill-from-github.py — Installs skills from GitHub
  • gh-address-comments/scripts/fetch_comments.py — Fetches PR comments
  • plugin-creator/scripts/create_basic_plugin.py — Scaffolds plugin directory
05

Prompts

OpenAI Skills (Official) — Prompts

Prompt 1: skill-installer — In-Session Discovery

File: skills/.system/skill-installer/SKILL.md Technique: Self-referential bootstrap with graceful fallback

---
name: skill-installer
description: Install Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos).
---

Install skills with the helper scripts.

## Communication

When listing skills, output approximately as follows:
"""
Skills from {repo}:
1. skill-1
2. skill-2 (already installed)
3. ...
Which ones would you like installed?
"""

After installing a skill, tell the user: "Restart Codex to pick up new skills."

Technique analysis: Prescribes the exact output format for skill listing (numbered list with installed annotations), making the agent's output predictable and parseable. The "Restart Codex" post-install message is baked into the skill instructions.


Prompt 2: openai-docs — MCP-Preferred Lookup with Fallback Chain

File: skills/.system/openai-docs/SKILL.md Technique: Priority-ordered tool selection with explicit fallback chain

## Workflow

1. Clarify whether the request is general docs lookup, model selection, a model-string upgrade, prompt-upgrade guidance, or broader API/provider migration.
2. For model-selection or upgrade requests, prefer current remote docs over bundled references when the user asks for latest/current/default guidance.

- Use `mcp__openaiDeveloperDocs__search_openai_docs` to find the most relevant doc pages.
- Use `mcp__openaiDeveloperDocs__fetch_openai_doc` to pull exact sections.
- Use `mcp__openaiDeveloperDocs__list_openai_docs` only when you need to browse without a clear query.

## If MCP server is missing

If MCP tools fail or no OpenAI docs resources are available:
1. Run the install command yourself: `codex mcp add openaiDeveloperDocs --url https://developers.openai.com/mcp`
2. If it fails due to permissions/sandboxing, immediately retry with escalated permissions...
3. Only if the escalated attempt fails, ask the user to run the install command.

Technique analysis: Three-tier fallback chain: MCP first → escalated permissions retry → user manual install. The model-preservation instruction ("if the user names a target model like 'migrate to GPT-5.4', keep that requested target even if latest-model.md names a newer model") is a sophisticated anti-hallucination guard.


Prompt 3: gh-address-comments — Selective Comment Processing

File: skills/.curated/gh-address-comments/SKILL.md Technique: Number-and-select approval gate

## 2) Ask the user for clarification
- Number all the review threads and comments and provide a short summary of what would be required to apply a fix for it
- Ask the user which numbered comments should be addressed

## 3) If user chooses comments
- Apply fixes for the selected comments

Technique analysis: Forces agent to enumerate and summarize before acting. The "number all review threads" step converts an ambiguous PR into a discrete selectable set, preventing the agent from assuming which comments to fix.


Prompt 4: plugin-creator — Explicit Normalization Rules

File: skills/.system/plugin-creator/SKILL.md Technique: Constraint specification with normalization rules

# Plugin Creator

## Quick Start

1. Run the scaffold script:

```bash
  # Plugin names are normalized to lower-case hyphen-case and must be <= 64 chars.
  # The generated folder and plugin.json name are always the same.
python3 .agents/skills/plugin-creator/scripts/create_basic_plugin.py <plugin-name>

**Technique analysis:** Embeds normalization rules directly in the instruction comment (`lower-case hyphen-case`, `<= 64 chars`) rather than in validation code, relying on the agent to self-enforce constraints.
09

Uniqueness

OpenAI Skills (Official) — Uniqueness

Differs from Seeds

No seed framework parallels openai/skills' position as an integration library with tiered skill management:

  • Unlike superpowers or agent-skills (workflow methodology), openai/skills is a collection of integration skills for external services (Figma, Vercel, GitHub, Sentry, Linear, Notion) with no development workflow enforcement.
  • Unlike anthropics/skills (reference gallery), openai/skills has a three-tier curation system (.system/.curated/.experimental) with an in-session discovery and install mechanism (skill-installer).
  • Unlike taskmaster-ai (Archetype 3 — MCP-anchored toolserver), openai/skills uses the Agent Skills markdown format rather than MCP tools.
  • The closest seed is agent-os (Archetype 4 — markdown scaffold) but openai/skills is far more operationally rich with deployment skills, scripted helpers, and the sandbox escalation model.

Positioning

The official Codex counterpart to Anthropic's official skills repository. Its key differentiators:

  1. Tiered curation.system/.curated/.experimental tiers give a quality signal that anthropics/skills lacks.
  2. In-session install$skill-installer is a bootstrap skill that enables agents to extend themselves during a session.
  3. Integration depth — 7 Figma skills, 4 Notion skills, 4 deployment targets (Vercel/Netlify/Cloudflare/Render), 3 security skills, and Sentry/Linear/Playwright/Jupyter integrations.
  4. Sandbox escalation model — explicit capability-gating referenced across multiple skills.

Most Interesting Component

skill-installer — the only self-extending mechanism in the batch. An agent that has skill-installer can bootstrap its own capabilities by listing and installing more skills within the same session.

Observable Failure Modes

  1. Codex-locked — sandbox model, $CODEX_HOME, and .codex-plugin/ structure all tie skills to Codex, limiting portability.
  2. No session enforcement — skills activate by description only; without session-start injection, agents may not discover relevant skills for a given task.
  3. Restart required — newly installed skills require Codex restart, breaking the session continuity that makes in-session install valuable.
  4. Per-skill licensing complexity — each skill has its own license; developers must check each skill's LICENSE.txt before use.

Explicit Antipatterns

  • "The skills at skills/.system are preinstalled, so no need to help users install those. If they ask, just explain this." (from skill-installer)
  • Git fallback for private repos: "tries HTTPS first, then SSH" (implicit anti-pattern: don't assume SSH)
04

Workflow

OpenAI Skills (Official) — Workflow

General Pattern

No enforced workflow. Skills are standalone capabilities invoked when the description matches the task. The only workflow-enforcement skill is define-goal (ask user to define clear goals before starting work).

Skill Discovery and Install Workflow

This is the most interesting workflow in the repo:

  1. User asks Codex to install a skill or list available skills
  2. $skill-installer (pre-installed system skill) activates
  3. Lists skills from skills/.curated via GitHub API, annotating already-installed ones
  4. User picks skill(s) to install
  5. install-skill-from-github.py fetches and installs to $CODEX_HOME/skills/<skill-name>
  6. User restarts Codex to load new skills

Install options:

  • Public repos: direct download
  • Auth failures: falls back to git sparse checkout
  • Private repos: uses existing git credentials or GITHUB_TOKEN

Plugin Creation Workflow (plugin-creator)

  1. Run scaffold script: python3 .agents/skills/plugin-creator/scripts/create_basic_plugin.py <plugin-name>
  2. Open .codex-plugin/plugin.json and fill [TODO: ...] placeholders
  3. Optionally add companion folders: --with-skills --with-hooks --with-scripts --with-assets --with-mcp --with-apps
  4. Generate marketplace entry with --with-marketplace

PR Comment Workflow (gh-address-comments)

  1. Verify gh authentication
  2. Run scripts/fetch_comments.py — prints all comments and review threads
  3. Number and summarize threads, ask user which to address
  4. Apply fixes for selected comments

Phase-to-Artifact Map

Phase Artifact
Skill install $CODEX_HOME/skills/<skill-name>/
Plugin scaffold .codex-plugin/plugin.json + optional dirs
PR comment fix Code changes + PR update
Deployment Deployed app on target platform

Approval Gates

Gate Type
User chooses which PR comments to address choice-list
User restarts Codex after skill install manual
User fills TODO placeholders in plugin.json file-review
06

Memory Context

OpenAI Skills (Official) — Memory & Context

State Storage

  • Type: File-based (skills installed to $CODEX_HOME/skills/)
  • Persistence: Global (skills persist across Codex sessions once installed)
  • Install path: $CODEX_HOME/skills/<skill-name>/ (defaults to ~/.codex/skills/)

Session Context

No session-start hook. System skills are pre-loaded at Codex launch. Curated/experimental skills are available after install + restart.

Cross-Session Handoff

Installed skills persist globally at $CODEX_HOME/skills/. The already installed annotation in skill-installer's list output reflects the current $CODEX_HOME/skills state.

Context Compaction

No explicit mechanism.

Live Documentation Loading

The openai-docs skill loads documentation via MCP tools or WebFetch — dynamic loading of current content:

  • mcp__openaiDeveloperDocs__search_openai_docs
  • mcp__openaiDeveloperDocs__fetch_openai_doc
  • Falls back to bundled references/latest-model.md

None.

Notable State Files

  • $CODEX_HOME/skills/ — installed skill directories
  • $CODEX_HOME/.agents/plugins/marketplace.json — marketplace registry for plugin ordering
07

Orchestration

OpenAI Skills (Official) — Orchestration

Multi-Agent Support

No — no subagent definitions, no spawn mechanisms.

Orchestration Pattern

None at the framework level. Individual skills may instruct agents to run parallel sub-tasks (e.g., skill-creator runs evals in background), but there is no framework-level coordination.

Isolation Mechanism

None — skills execute in the Codex session's current working directory.

Multi-Model

No — Codex single model. The openai-docs skill references gpt-oss-120b and gpt-oss-20b in its product snapshot but does not route to multiple models.

Execution Mode

Event-driven — skills activate when description matches the user's task.

Consensus Mechanism

None.

Prompt Chaining

Weak — define-goal outputs a goal that subsequent skills could use, but this is not enforced or formalized.

Auto-Validators

No framework-level auto-validation. Some skills include explicit validation steps (e.g., gh-fix-ci analyzes CI failures and verifies fixes), but these are manual workflow steps within the skill, not hook-triggered.

Git Automation

  • gh-address-comments: guides agent to apply fixes (code changes), but does not commit automatically
  • gh-fix-ci: guides agent to fix CI (code changes), but commit is manual

Sandbox / Permissions

Multiple skills note "run all gh commands with elevated network access" and "request escalation when running them in the sandbox" — indicating Codex has a sandboxed execution model where some operations require explicit permission escalation.

08

Ui Cli Surface

OpenAI Skills (Official) — UI & CLI Surface

Dedicated CLI Binary

None — no CLI binary in this repo. Skills are consumed by the Codex agent runtime.

Codex-Specific Install

The $skill-installer system skill serves as an in-session install mechanism. Users interact with it via natural language:

  • "List installable skills"
  • "Install gh-address-comments"
  • "Install the create-plan skill from the experimental folder"

Local Web Dashboard

None.

IDE Integration

None explicitly. Codex may run in VS Code or other environments, but this repo ships no IDE-specific config.

Cross-Tool Portability

Low — optimized for Codex (OpenAI's agent). The Agent Skills standard format (SKILL.md) is theoretically portable, but:

  • Sandbox/escalation model is Codex-specific
  • $CODEX_HOME path is Codex-specific
  • mcp__openaiDeveloperDocs__* tool names are Codex-specific
  • Plugin creator scaffold paths (.codex-plugin/) are Codex-specific

Observability

No audit log, no replay capability.

Sandbox Model

Multiple skills reference Codex's sandbox model:

  • "request escalation when running in the sandbox"
  • "retry with escalated permissions"
  • sandbox_permissions=require_escalated

This indicates Codex has a capability-gated execution model where network access, GitHub auth, and other privileged operations require explicit escalation approval — a security model not present in Claude Code.

Related frameworks

same archetype · same primary tool · same memory type

Superpowers ★ 207k

Enforces spec-first, TDD, and subagent-reviewed development as mandatory automatic workflows rather than optional practices.

Anthropic Skills (Official) ★ 141k

Official Anthropic reference gallery for the Agent Skills standard, spanning creative, technical, and enterprise capability…

daymade/claude-code-skills ★ 1.1k

52 production-hardened skills with emphasis on skill craftsmanship — a fork of Anthropic's official skill-creator with security…

Superpowers Lab ★ 344

Experimental skill incubator extending Superpowers with interactive CLI automation, on-demand MCP discovery, semantic…

openspec-skills (chyiiiiiiiiiiii) ★ 6

Ports the OpenSpec proposal/apply/archive SDD lifecycle into three plain SKILL.md files requiring zero runtime dependencies.

Anthropic Skills — IP Guard Delta Report ★ 141k

No ip-guard skill found in the anthropics/skills repo; this entry is a delta report tracking its absence.