OpenAI Codex skills

Codex skills for repeatable coding and agent workflows.

A Codex skill is a folder of instructions, scripts, and references for one repeatable job. Install a focused workflow, inspect what it can touch, and prove it on a small task before trusting it with real work.

Browse 3 installable Codex skills Read the official Codex documentation

User-level skills: $HOME/.agents/skills
Repository skills: $REPO_ROOT/.agents/skills
Required file: SKILL.md

Understand the format

See what Codex loads and what belongs outside SKILL.md.

Install one workflow

Use the current user or repository skill location.

Verify before trusting

Check scope, secrets, failure behavior, and clean removal.

How Codex skills work

Keep the trigger small. Load the full workflow only when it fits.

Codex uses progressive disclosure. It first sees each skill's name, description, and path. It reads the full instructions only after selecting that skill, which keeps unrelated workflow detail out of every task.

Required

SKILL.md

Defines the skill name, the trigger description Codex sees, and the workflow Codex loads when the skill is selected.

Optional

scripts/

Holds deterministic commands for work that should not depend on the model rewriting shell or application logic every run.

Optional

references/

Keeps API notes, decision tables, schemas, and longer operating context out of the main instructions until needed.

Optional

assets/

Stores templates and files the workflow copies or transforms instead of loading them into model context.

AGENTS.md

Rules for the repository.

Use AGENTS.md for standing project guidance Codex should read before every task: package manager, architecture, verification commands, boundaries, and conventions.

Read the official AGENTS.md guide

SKILL.md

A workflow for one job.

Use a skill for work Codex should load only when the task matches: releasing a package, reviewing a migration, importing content, or producing a verified handoff.

Rule of thumb: AGENTS.md explains how this repo works. A skill explains how to complete this repeatable job.

MCP server

A connector to external tools or data.

MCP expands what Codex can call. A skill defines when and how to complete a repeatable job.

Codex CLI

The coding-agent runtime.

Install Codex itself before adding skills, then choose only the workflows your work actually needs.

Open the Codex CLI guide

First-party Codex skills

Start with a workflow you can inspect.

These BuildLeanSaaS skills expose their instructions and bundled scripts. Read the source, confirm the required credentials, then install only the capability you need.

Monitor trusted release sources

Release Source Capture

Free Codex/Hermes skill for monitoring GitHub releases and RSS feeds, deduping changes locally, and emitting JSON events for sourced research briefs.

Verification: Dry-run emits normalized JSON without updating saved state.

Inspect skill

Capture saved research into an inbox

X Bookmark Capture Sync

An installable OpenClaw/Codex-style skill that captures your X bookmarks on a cron job and turns them into raw GitHub Issues labeled inbox.

Verification: Dry-run previews the capture plan before writing downstream records.

Inspect skill

Turn raw captures into shaped work

Linear X Bookmark Agent Triage

An installable Hermes/Codex-style skill for reviewing Linear issues created by X Bookmark Capture Sync and turning each capture into a Shape Up Linear task.

Verification: Dry-run previews routing decisions before changing the work queue.

Inspect skill

Install example

Install for your user, then start a fresh Codex session.

User skills belong under $HOME/.agents/skills. Repository skills belong under .agents/skills inside the relevant repo. A new session gives Codex a clean chance to discover the skill.

mkdir -p "$HOME/.agents/skills/bookmark-capture-sync/scripts"
curl -fsSL https://www.buildleansaas.com/api/skills/x/bookmark-capture-sync/files/SKILL.md \
  -o "$HOME/.agents/skills/bookmark-capture-sync/SKILL.md"
curl -fsSL https://www.buildleansaas.com/api/skills/x/bookmark-capture-sync/files/scripts/bookmark-capture-sync.mjs \
  -o "$HOME/.agents/skills/bookmark-capture-sync/scripts/bookmark-capture-sync.mjs"
chmod +x "$HOME/.agents/skills/bookmark-capture-sync/scripts/bookmark-capture-sync.mjs"

Build your own

One trigger. One outcome. One verification path.

Start with the smallest workflow you repeat. Put decisions and ordered steps in SKILL.md. Move deterministic work into scripts and longer facts into references.

my-workflow/
├── SKILL.md
├── scripts/
│   └── verify.mjs
├── references/
│   └── api-notes.md
└── assets/
    └── output-template.md

Evaluation checklist

Do not judge a skill by its README.

The description names a specific trigger and outcome.
The workflow states which files, services, or commands it may touch.
Secrets stay in environment variables or approved secret stores.
Scripts fail clearly instead of inventing plausible output.
The skill includes a focused verification step.
A clean-profile smoke test proves installation and discovery.
Removal does not damage files that existed before installation.

From skill to operating system

Connect reusable skills to an always-on agent workflow.

The free Always-On Agents course covers persistent context, scheduled jobs, verification receipts, and safe handoffs between the operator and deeper coding workers.

Open the free course