Best Places to Find Codex Skills in 2026

The Remote OpenClaw Codex skills directory is the best place to find skills for OpenAI Codex in 2026, followed by skills.sh, agentskills.io, aitmpl.com, and GitHub search. Codex adopted the open Agent Skills standard, so both Codex-focused directories and cross-agent skill catalogs are legitimate sources, and this roundup ranks five of them by catalog size, Codex-native install support, and curation quality, each verified live in early July 2026.

Disclosure: Remote OpenClaw is our own directory, so we rank it first with an obvious interest. Every claim behind that ranking is verifiable on our public stats page, and every competitor gets a fair entry with real strengths and one honest limitation.

How Codex Skills Work

OpenAI Codex supports agent skills built on the open Agent Skills standard, per the official Codex skills documentation. A skill is a folder with a SKILL.md file, and Codex discovers skills from .agents/skills in your repository, $HOME/.agents/skills at the user level, and /etc/codex/skills for admin-managed installs.

You invoke a skill explicitly with the /skills command or a $ mention, or Codex selects one implicitly when your task matches its description. Codex also ships a $skill-installer skill that pulls curated skills into your local setup. Because the format is the same SKILL.md used by Claude Code and other agents, the practical question is not "which directories are Codex directories" but "which directories index skills Codex can read," which is what this ranking answers. (For other runtimes, see the companion roundups on the best places to find Claude Code skills and the cross-agent best agent skills directories.)

1. Remote OpenClaw Codex Directory (Best Overall)

The Remote OpenClaw Codex skills directory is a dedicated hub for Codex that lists official and community skills in one place, with each listing linking to the source so you can read the SKILL.md before installing. If you want a shortlist rather than a catalog, the companion best Codex skills page ranks the picks worth installing first.

The hub is part of the wider Remote OpenClaw index, which tracked 13,870+ MCP servers, 4,384+ agent skills, and 3,687+ plugins across five agent runtimes (Claude Code, Codex, OpenClaw, Hermes, plus Cursor and other MCP clients) as of early July 2026. Three verifiable reasons it ranks first:

• Free with no signup. Browse, search, and copy install details without an account.
• Public analytics. Catalog counts update daily on the stats page, so the size claim is auditable rather than marketing copy.
• Terminal access. npx remoteopenclaw search <query> queries the same index from your shell or from inside an agent via the open-source MCP server.

The honest caveat: it is our directory, which is why the ranking rests on public numbers instead of adjectives.

2. skills.sh (Best Codex-Native Install)

skills.sh is Vercel's agent skills leaderboard, and it gives Codex users the smoothest install path of any source here: npx skills add <owner/repo> places skills directly into supported agents, and Codex is one of its 22 named targets alongside Claude Code, Cursor, and GitHub Copilot. It is also the only directory on this list with public install data; its top skill, find-skills from vercel-labs, showed 2.3M installs as of July 2026, and trending filters show what developers adopted in the last 24 hours.

One limitation: skills.sh ranks by popularity and does not publish vetting or security review criteria, so a high install count is a social signal, not a safety guarantee.

3. agentskills.io (The Standard Itself)

agentskills.io is the home of the Agent Skills specification that Codex skills are built on, originally developed by Anthropic and released as an open standard. For Codex specifically it offers three things: the full SKILL.md format specification, a quickstart for writing your first skill, and a client showcase that lists OpenAI Codex as an official adopter with a direct link to the Codex skills docs, alongside dozens of other agents from Cursor to Gemini CLI.

It is the right first stop when you plan to write or publish skills rather than just consume them, since skills that follow the spec exactly will work across every adopting agent.

One limitation: agentskills.io is a specification site, not a large browsable catalog; it points you to the format and the ecosystem but hosts no install counts or per-skill listings at meaningful scale.

4. aitmpl.com (Biggest Component Library)

aitmpl.com is the web front end for davila7's claude-code-templates project, which advertises 1,000+ ready-to-use components: skills, agents, commands, hooks, MCPs, and plugins. Its skills are written in the same SKILL.md format Codex reads, so the catalog is a real Codex resource even though the branding says Claude Code, and a stack-builder interface lets you assemble a set of components before downloading.

One limitation: the site is built around Claude Code, so there is no Codex install path; you copy a skill folder into .agents/skills yourself and check each skill for Claude-specific tooling (hooks and slash commands, for example, will not carry over).

5. GitHub Search (Biggest Raw Pool)

GitHub is the largest unfiltered source of Codex-compatible skills, because every skill on every directory above ultimately lives in a GitHub repository. Searching code for path:SKILL.md codex surfaces skills that mention Codex directly, the openai/codex repository documents the loader behavior, and the agent-skills topic aggregates cross-agent collections that no single directory has fully indexed.

One limitation: there is zero curation and no install tooling; star counts are the only quality signal, and you must read every SKILL.md and bundled script before giving it to an agent that runs shell commands.

Comparison Table

Here is how the five sources compare for Codex specifically, verified in July 2026.

When Not to Install a Skill

Skip third-party skills when the capability is already native to Codex or when the skill needs credentials you would not paste into a prompt. Codex loads skill descriptions into limited context (skill lists are capped at 8,000 characters), so hoarding dozens of marginal skills degrades discovery of the ones you actually use; keep the repo-level .agents/skills folder small and purposeful.

Also remember that a skill is executable guidance for an agent with shell access. None of the free directories above audits skills the way a package registry audits packages, so treat an unread SKILL.md the way you would treat an unread install script. If you are still choosing between agents rather than choosing skills, our Claude Code vs Codex vs Cursor comparison covers how their skill systems differ.

Related Guides

• OpenAI Codex Guide
• Codex CLI MCP: How to Add MCP Servers to Codex
• Claude Code vs Codex vs Cursor Comparison
• OpenClaw Codex Web Search Guide