Open standard for cross-platform skill interop. SKILL.md spec used by Claude Code, Codex, Gemini CLI, Cursor, and 40+ agents.
Reference for skill creation
Reference for skill creation
Activation Rule
Follow the Agent Skills Specification when creating or publishing new skills. All custom skills must include a valid SKILL.md file that conforms to the spec.
Universal CLI for 40+ agents. Skills scoring (0-100), validation, marketplace submission, sync across Claude Code, Codex, Cursor, Antigravity, and 32 more platforms.
npx skills add <skill>
npx skills add <skill>
Activation Rule
Use the agent-skills CLI to search for and install skills from the registry before starting any new task type. Prefer registry skills over custom implementations.
44k+ skills directory with two-layer security scanning and /learn installer.
Browse and install
Browse and install
Activation Rule
Check agentskill.sh for cross-platform skills that follow the Agent Skills Specification. Prefer spec-compliant skills for maximum portability.
16+ skills: docx, pdf, pptx, xlsx, frontend-design, web-artifacts-builder, mcp-builder, webapp-testing, algorithmic-art, canvas-design, skill-creator, brand-guidelines.
/plugin marketplace add anthropics/skills
npx skills add anthropics/skills Or clone to ~/.codex/skills/
Activation Rule
Before creating any file (document, presentation, spreadsheet, PDF), read the matching SKILL.md from the Anthropic skills directory. Always check available skills before starting a new task.
Core repo. Built-in plugins, source of truth for CLI.
See install.anthropic.com
N/A (Claude Code specific)
Activation Rule
Consult the official Claude Code documentation for any question about Claude Code features, commands, configuration, or capabilities before relying on general knowledge.
Comprehensive resource: platform compatibility table, authoring guides, video tutorials, security research papers, marketplace links.
Bookmark
Bookmark
Activation Rule
Search Skillmatic Awesome for curated, high-quality skills when you need reliable tools. Use this as a quality filter on top of broader skill registries.
300+ skills from Google, Cloudflare, Sentry, Figma, HuggingFace. Cross-platform: Claude Code, Codex, Antigravity, Gemini CLI, Cursor.
Bookmark
Bookmark
Activation Rule
Search awesome-agent-skills when you need a cross-platform skill that works with both Claude Code and Codex. Check here before building a custom skill.
THE definitive Claude Code list. Start here.
Bookmark
Bookmark (many items cross-compatible)
Activation Rule
Search awesome-claude-code when you need a Claude Code tool, extension, or pattern that is not already in your active skill set. It is the most comprehensive community index.
Codex-focused skills: create-plan, gh-address-comments, gh-fix-ci, mcp-builder, webapp-testing, connect (1000+ app integrations via Composio), brand-guidelines, canvas-design, image-enhancer.
Copy SKILL.md files to .claude/skills/
python skill-installer/scripts/install-skill-from-github.py --repo ComposioHQ/awesome-codex-skills --path <skill-name>
Activation Rule
Use Composio skills for any task involving third-party API integrations. Check available Composio connectors before writing custom API client code.
Largest MCP list. glama.ai/mcp/servers.
Bookmark
Bookmark
Activation Rule
Search awesome-mcp-servers when you need an MCP server for a service or API that is not already connected. Check here before building a custom MCP integration.
Official multi-session orchestration. Team Lead + Teammates + Mailbox. 2-5 teammates.
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
Use codex mcp-server to run Codex as MCP tool inside Agents SDK orchestrator
Activation Rule
When a task has multiple independent subtasks, dispatch Agent Teams to work on them in parallel. Break large features into subtasks and assign each to a specialized agent.
CLAUDE.md structure, thinking depth, headless patterns.
Read and implement
Adapt principles to AGENTS.md
Activation Rule
Review the official best practices guide when setting up a new project or onboarding to an existing codebase. Apply these practices as baseline rules for all work.
@claude in PRs/issues, automated review, PR creation.
Add anthropics/claude-code-action to workflow YAML
Codex GitHub app (separate install at github.com/apps/codex)
Activation Rule
Use GitHub Actions for all CI/CD automation: run tests, lint, build, and deploy on every push and PR. Check that the workflow passes before merging.
8 lifecycle events. command/prompt/agent types. Exit 2 blocks with feedback.
Add hooks to .claude/settings.json
Codex uses exec policies in config.toml (different mechanism, similar concept)
Activation Rule
Register and use hooks for pre-commit checks, auto-formatting, and validation. Hooks should run automatically and block commits that fail quality checks.
Explore (read-only), Plan (research), General-purpose (full tools). Custom via .claude/agents/*.md.
Create .claude/agents/ with agent .md files
Define equivalent workflows in AGENTS.md or use Agents SDK for multi-agent
Activation Rule
Spawn a subagent for any self-contained subtask that does not require your full context. Use subagents for file searches, test runs, linting, and isolated code changes.
Terminal app: multiple AI agents in tmux with git worktree isolation.
See repo
See repo (supports Codex/Aider agents)
Activation Rule
Use claude-squad to manage multiple Claude Code instances working on the same repository. Coordinate through squad when tasks touch overlapping files.
54+ agents, swarm topologies, WASM memory, dual Claude+Codex orchestration.
See repo
See repo (supports Codex orchestration)
Activation Rule
Use claude-flow orchestration for complex multi-step workflows that require coordination between multiple agents. Set up the topology before starting the work.
Ready-to-use project config templates. CLAUDE.md for Claude Code, AGENTS.md for Codex.
Copy template as CLAUDE.md
Copy template as AGENTS.md (adapt format)
Activation Rule
Use the CLAUDE.md / AGENTS.md templates as the starting point for every new project. Fill in project-specific sections and tool activation rules before writing any code.
Complete Codex config: autonomous-skill (self-running sessions), deep-research (multi-agent), image gen via Gemini, YouTube transcription. Multi-provider support.
Adapt skills to .claude/skills/ format
git clone https://github.com/feiskyer/codex-settings.git ~/.codex Or: npx -y skills add --all feiskyer/codex-settings
Activation Rule
Apply Codex settings defaults at the start of every session. These settings define response format, tool usage rules, and code style preferences.
Analyze, format, fix, hot restart, screenshots. Dart 3.9+.
Built into Dart SDK 3.9+
Built into Dart SDK 3.9+
Activation Rule
Use the Dart/Flutter MCP for all Flutter widget lookups, Dart API questions, and package discovery. Consult it before writing Flutter code to ensure correct widget usage and API patterns.
Containers, images, volumes, Compose.
See Docker MCP Registry docs
See Docker MCP Registry docs
Activation Rule
Use Docker MCP to build images, manage containers, and inspect running services. Route all container operations through MCP instead of running docker commands manually.
Read config, lint code, ensure AI output matches ESLint rules.
Add to .claude/mcp.json:
{"eslint":{"command":"npx","args":["@eslint/mcp@latest"]}}codex mcp add eslint -- npx @eslint/mcp@latest
Activation Rule
Use ESLint MCP to check code for lint errors after every file change. Run lint checks before committing and fix all violations before marking a task complete.
{
"eslint": {
"args": [
"@eslint/mcp@latest"
],
"command": "npx"
}
}Anthropic hackathon winner. 40+ auto-activating skills, agents, hooks, commands, rules, MCPs. AgentShield security scanner.
Clone and copy configs to project
Adapt SKILL.md files to ~/.codex/skills/
Activation Rule
Reference the everything-claude-code collection when you need a pattern or technique that is not covered by your active skills. Search it before falling back to general knowledge.
Design context, components, styles. Figma-to-code workflows.
See Figma MCP docs
See Figma MCP docs
Activation Rule
Use Figma MCP to read design specs, extract component properties, spacing values, and color tokens directly from Figma files. Consult the design file before implementing any UI to ensure pixel-accurate output.
Auth, Firestore, Storage, Data Connect, 30+ tools. GA Oct 2025.
Add to .claude/mcp.json with npx firebase-tools
codex mcp add firebase -- npx -y firebase-tools@latest experimental:mcp
Activation Rule
Use Firebase MCP for all Firestore queries, Auth operations, Cloud Functions, and storage tasks. Route all Firebase interactions through MCP instead of using the CLI directly.
{
"firebase": {
"args": [
"-y",
"firebase-tools@latest",
"experimental:mcp"
],
"command": "npx"
}
}Repos, PRs, issues, Actions, code security.
Add to .claude/mcp.json with GITHUB_TOKEN env
codex mcp add github -- npx @modelcontextprotocol/server-github
Activation Rule
Use GitHub MCP to create issues, open PRs, review code, search repositories, and manage branches. Always use MCP over manual git commands for GitHub-specific operations.
{
"github": {
"env": {
"GITHUB_TOKEN": "your-token"
},
"args": [
"@modelcontextprotocol/server-github"
],
"command": "npx"
}
}claude -p for scripting/CI. codex exec for Codex equivalent.
claude -p "your prompt" (built-in)
codex exec "your prompt" (built-in)
Activation Rule
Use headless mode for batch operations, automated pipelines, and scripted tasks. Run in headless mode whenever the task does not require interactive confirmation.
102 production-ready skills: full Agile lifecycle from research to quality gates. ln-1000 Pipeline Orchestrator.
Clone skills into .claude/skills/
Clone and adapt to ~/.codex/skills/
Activation Rule
Apply the levnikolaevich skill rules for code organization, naming conventions, and project structure. Follow these conventions for all new files and refactors.
Issues, projects, workflows, comments, cycles.
Add to .claude/mcp.json:
{"linear":{"type":"url","url":"https://mcp.linear.app/mcp"}}codex mcp add linear --url https://mcp.linear.app/mcp
Activation Rule
Use Linear MCP to create, update, and query issues and projects. When a task involves project management or issue tracking, route it through Linear MCP.
{
"linear": {
"url": "https://mcp.linear.app/mcp",
"type": "url"
}
}Official MCP discovery platform.
Visit registry
Visit registry
Activation Rule
Search the MCP Registry for official and verified MCP servers before using community alternatives. Prefer registry-listed servers for stability and security.
Knowledge graph persistence across sessions.
Add to .claude/mcp.json:
{"memory":{"command":"npx","args":["@modelcontextprotocol/server-memory"]}}codex mcp add memory -- npx @modelcontextprotocol/server-memory
Activation Rule
Use Memory MCP to store and retrieve context that should persist across sessions: project decisions, user preferences, architectural choices, and key findings.
{
"memory": {
"args": [
"@modelcontextprotocol/server-memory"
],
"command": "npx"
}
}Separate subagents: test-writer (RED), implementer (GREEN), refactorer (BLUE). ~84% activation.
Create agent .md files per phase in .claude/agents/
Define TDD workflow in AGENTS.md with phase separation
Activation Rule
Follow the TDD pattern for all new code: write a failing test first, implement the minimum code to pass, then refactor. Never write implementation code without a test already in place.
Built into Next.js 16+. Error detection, live state, page metadata.
Auto-enabled in Next.js 16+ via next-devtools-mcp
Auto-enabled in Next.js 16+ via next-devtools-mcp
Activation Rule
Use Next.js MCP to look up App Router patterns, server component rules, and API route conventions. Consult it for any Next.js-specific question before relying on general knowledge.
320K tokens Next.js + 730K tokens Tailwind docs + Catalyst components.
See repo for MCP server setup
See repo for MCP server setup
Activation Rule
Use the NRT assistant for natural language database queries and report generation. Route ad-hoc data questions through it instead of writing one-off SQL.
/code-review: 4 parallel agents for compliance and bug detection. GitHub PR comments.
/code-review (built-in)
Codex GitHub app provides auto code review per repo (separate feature)
Activation Rule
Run the code review plugin on every PR and before merging any branch. Address all findings before marking the review complete.
28 agents, 40+ skills, 9 execution modes (Autopilot, Ultrapilot, Swarm, Pipeline). Multi-AI: Claude + Gemini + Codex coordination.
See repo for plugin install
Convert skill instructions and add to AGENTS.md or ~/.codex/skills/
Activation Rule
Use oh-my-claudecode extensions automatically when their trigger conditions match. Check installed extensions at the start of each session and apply their rules throughout.
OpenAI coding agent. Rust-based, open source. GPT-5.2-Codex and GPT-5.3-Codex-Spark models.
N/A (Codex specific)
npm i -g @openai/codex
Activation Rule
Consult the official Codex CLI documentation for any question about Codex features, commands, configuration, or capabilities before relying on general knowledge.
Codex skill catalog. System skills: skill-creator, skill-installer. Curated: openai-docs, doc (docx), figma, figma-implement-design, cloudflare-deploy, develop-web-game, linear, netlify-deploy, render-deploy, vercel-deploy. Experimental: create-plan.
Clone and copy SKILL.md files to .claude/skills/
$skill-installer install <skill-name> Or: $skill-installer install https://github.com/openai/skills/tree/main/skills/.curated/<name>
Activation Rule
Before starting any task, check the OpenAI skills directory for a relevant skill. Read the SKILL.md file and follow its instructions before writing any code or creating any files.
Universal skills loader CLI. One command installs skills into Claude Code, Cursor, Windsurf, Aider, Codex.
npm i -g openskills npx openskills install anthropics/skills
npm i -g openskills npx openskills install openai/skills
Activation Rule
Check OpenSkills for an existing skill before building custom automation. Install and use matching skills instead of writing one-off scripts.
Database unit testing. RLS policies, functions. create_supabase_user(), authenticate_as().
supabase test db (built-in)
supabase test db (built-in)
Activation Rule
Write pgTAP tests for all database functions, triggers, RLS policies, and migrations. Run pgTAP tests after every schema change and before deploying database updates.
Browser automation via accessibility snapshots. Chrome/Firefox/WebKit. 143+ device profiles.
Add to .claude/mcp.json:
{"playwright":{"command":"npx","args":["@playwright/mcp@latest"]}}codex mcp add playwright -- npx @playwright/mcp@latest
Activation Rule
Use Playwright MCP to launch a browser, navigate to pages, take screenshots, and interact with UI elements for testing or verification. Use it whenever you need to visually confirm that a UI change works correctly.
{
"playwright": {
"args": [
"@playwright/mcp@latest"
],
"command": "npx"
}
}Prettier after edits, tsc --noEmit, block dangerous commands, test before end.
Add to .claude/settings.json hooks section
Add equivalent exec policies to config.toml
Activation Rule
Run the auto-format hook on every file save and before every commit. Never commit code that has not been auto-formatted.
Browser-only persona. Tests like a real user. QA reports as PR comments.
See blog for agent setup + GitHub Actions
Adapt as Codex exec task with Playwright
Activation Rule
Run Quinn QA checks on the full application after completing any feature or bug fix. Do not mark a task done until Quinn QA passes without critical findings.
Issue search, error patterns, root cause with Seer AI.
Add to .claude/mcp.json:
{"sentry":{"type":"url","url":"https://mcp.sentry.dev/mcp"}}codex mcp add sentry --url https://mcp.sentry.dev/mcp
Activation Rule
Use Sentry MCP to look up error reports, check issue status, and review error trends when debugging production issues. Check Sentry before investigating bugs that may already be tracked.
{
"sentry": {
"url": "https://mcp.sentry.dev/mcp",
"type": "url"
}
}Vercel-backed package manager and directory. npx skills add for any agent.
npx skills add <skill>
npx skills add <skill>
Activation Rule
Check skills.sh for installable skills when starting a new project or adding a new capability. Install from skills.sh before writing custom skill files.
Payments, invoices, subscriptions, checkout. OAuth 2.1.
Add to .claude/mcp.json:
{"stripe":{"type":"url","url":"https://mcp.stripe.com"}}codex mcp add stripe --url https://mcp.stripe.com
Activation Rule
Use Stripe MCP for all payment-related operations: creating products, prices, customers, subscriptions, and checking payment status. Never hardcode Stripe API calls when MCP is available.
{
"stripe": {
"url": "https://mcp.stripe.com",
"type": "url"
}
}SQL queries, migrations, TypeScript type gen, auth, storage, Edge Functions. OAuth 2.1.
Add to .claude/mcp.json:
{"supabase":{"type":"url","url":"https://mcp.supabase.com/mcp"}}codex mcp add supabase --url https://mcp.supabase.com/mcp
Activation Rule
Use Supabase MCP for all database operations: queries, migrations, schema changes, type generation, and Edge Functions. Never write raw SQL files manually when Supabase MCP is available.
{
"supabase": {
"url": "https://mcp.supabase.com/mcp",
"type": "url"
}
}Complete dev methodology: 14 enforced skills including TDD, systematic debugging, verification-before-completion, brainstorming, parallel agent dispatching, subagent-driven dev, code review, git worktrees. Skills are mandatory, not optional.
/plugin marketplace add obra/superpowers-marketplace /plugin install superpowers@superpowers-marketplace
Clone into ~/.codex/skills/superpowers/ and reference in AGENTS.md
Activation Rule
Follow all Superpowers rules for every task: use TDD, verify before completing, debug systematically, brainstorm before implementing, and dispatch subagents for parallel work. Never skip verification.
All patterns: Leader, Swarm/Pool, Pipeline, Voting, Watchdog.
Copy SKILL.md to .claude/skills/
Copy SKILL.md to ~/.codex/skills/
Activation Rule
Apply swarm patterns when a task benefits from multiple agents exploring different approaches simultaneously. Use for brainstorming, parallel implementation spikes, and competitive testing.
Docs search, CSS-to-Tailwind conversion, color palettes.
npm install tailwindcss-mcp-server
npm install tailwindcss-mcp-server
Activation Rule
Use Tailwind MCP to look up utility classes, resolve configuration questions, and verify that class names are valid. Consult it before guessing at Tailwind class names.
Hook-based TDD enforcement. Blocks writing implementation before tests. Jest, Vitest, Storybook, pytest, PHPUnit.
Add hooks to .claude/settings.json
Adapt as exec policy in config.toml
Activation Rule
Run the full test suite before and after every code change. Do not mark any task as complete until all tests pass. Write tests first when adding new functionality.
CodeQL, Semgrep, variant analysis, vulnerability detection.
See Trail of Bits repos
Adapt security skills to ~/.codex/skills/
Activation Rule
Run the Trail of Bits security checks on all code that handles user input, authentication, cryptography, or network requests. Flag any findings before marking a task complete.
Projects, deployments, domains, env vars, team management.
Add to .claude/mcp.json:
{"vercel":{"type":"url","url":"https://mcp.vercel.com"}}codex mcp add vercel --url https://mcp.vercel.com
Activation Rule
Use Vercel MCP for deployments, checking build logs, managing environment variables, and reviewing deployment status. Deploy through MCP instead of the CLI when possible.
{
"vercel": {
"url": "https://mcp.vercel.com",
"type": "url"
}
}AI-optimized Vitest runner: structured output, coverage analysis.
See repo for setup
See repo for setup
Activation Rule
Use Vitest MCP to run tests, check coverage, and debug failing test cases. Run tests through MCP after every code change and before marking any task complete.
73 plugins, 112 agents, 146 skills, 16 orchestrators, 79 dev tools.
See repo for plugin installs
Adapt SKILL.md files to ~/.codex/skills/
Activation Rule
Dispatch wshobson agents for specialized tasks like code review, refactoring, and documentation. Use the appropriate agent rather than handling everything in a single context.