topskills

Codex guide

Best agent Skills for Codex CLI

The best Skills for Codex CLI are frontend-design for UI work, superpowers for planned multi-step engineering, and Trail of Bits' security Skills for review — all verified through the SKILL.md path Codex adopted in December 2025. 51 of the 191 Skills we catalogue are verified on Codex; because Codex is a terminal agent with real repo access, the review, testing, and infrastructure Skills carry their full capability here rather than a reduced one.

Updated August 26, 2026 · first published August 19, 2026

Install first

frontend-design

Verified on Codex and the highest-scoring Skill we catalogue. It substitutes a committed set of design decisions for the design file you do not have, which is the common case when generating UI from a terminal.

Best when: Generating a page or component from scratch with no design system to inherit.

Install model

How Skills install on Codex

codex skills add github.com/anthropics/skills/skills/frontend-design
  1. 1Copy the Codex install command from the Skill page — Codex takes a repo-plus-path form rather than a --skill flag.
  2. 2Run it from the project you want the Skill available in.
  3. 3Start a new Codex session so the Skill frontmatter is read.
  4. 4Trigger the Skill with a matching task and confirm the behaviour changed.

Codex adopted the SKILL.md spec directly, so a Skill folder written for Claude Code is the same folder here — the difference is the install path form (repo plus path to the skill directory) rather than the file format. The three-tier loading model is the same: metadata always, instructions on trigger, bundled files on demand.

Ranked

The shortlist

Ordered by what to install first for the common case, not by raw score. Every entry here carries verified Codex compatibility — open the skill page for the composite score, licence, and full install commands.

SkillSourceScoreBest when
frontend-designOfficial92Generating a page or component from scratch with no design system to inherit.
superpowersCommunity63Work that spans sessions and needs the plan to survive between them.
trail-of-bitsVerified49Reviewing code that touches auth, untrusted input, secrets, or payments.
debugOfficial51An intermittent or unexplained failure, before you start changing code.
paddle-agent-skillsVerified53Selling internationally where VAT and sales-tax handling is the actual problem, not the checkout form.
composioVerified55Connecting an agent to several third-party services where building each integration separately is the bottleneck.

See every skill verified on this agent, ranked by composite score, on the Codex page.

Why these

Pick by pick

  1. 1
    frontend-design

    Verified on Codex and the highest-scoring Skill we catalogue. It substitutes a committed set of design decisions for the design file you do not have, which is the common case when generating UI from a terminal.

  2. 2
    superpowers

    A full engineering pipeline with TDD enforced at each stage. Terminal agents are where long-running, multi-file work actually happens, so the plan-preserving Skill matters more here than in an editor.

  3. 3
    trail-of-bits

    Firm-grade security review Skills, Apache-2.0. Codex has real repo access, which is precisely the setting where a severity-tagged security pass over the whole diff beats inline suggestions.

  4. 4
    debug

    Replaces speculative edits with a hypothesis-and-test loop. Cheap, read-write, and verified on Codex — it changes agent behaviour on exactly the failures that burn the most session time.

  5. 5
    paddle-agent-skills

    Published by Paddle, covering merchant-of-record billing — the tax and compliance side generic payment advice skips entirely. One of the higher-scoring vendor Skills verified on Codex.

  6. 6
    composio

    One Apache-2.0 Skill that routes to a large catalogue of SaaS integrations, instead of wiring each one by hand. Verified across all seven agent targets, Codex included — the broadest compatibility of any integration Skill we list.

Limits

What this agent cannot do

Codex verification is narrower than Claude Code or Cursor: 51 of 191 Skills. That is a testing-coverage number, not a capability judgement — several vendor Skills in APIs, databases, and email have been verified on Claude Code and Cursor but not yet on Codex, and they are absent from these lists by policy rather than because they failed. The known real gaps are Skills built around Claude.ai upload flows or Claude API skill_id references, which have no Codex equivalent. Where a Skill you want is missing, the category page shows the full set with per-agent badges.

FAQ

Common questions

Does Codex CLI use the same SKILL.md format as Claude Code?

Yes — Codex adopted the spec directly, so the same folder works in both. Only the install command form differs: Codex takes the repo plus the path to the skill directory, rather than a repo plus a --skill flag.

Why are fewer Skills listed for Codex than for Cursor?

Because we list only what was tested. Verification is per Skill per agent and it takes real work, so coverage grows over time rather than arriving complete. A Skill missing from the Codex list has usually not been tested there yet.

Can Codex run scripts bundled inside a Skill?

Yes — Codex is a terminal agent, so bundled scripts are executable in the same way as in Claude Code. That is what makes the document and testing Skills worth installing here: the deterministic script does the work, not a fresh improvisation each time.

Do I need different Skills for Codex than for ChatGPT?

They are different surfaces with different capabilities. Codex CLI runs locally with filesystem and shell access, so it supports the full range including action-taking Skills. Judge each Skill by the surface it will run on rather than by the vendor.

Other agents

Same question, different agent

New to Skills? What is a Claude skill · How to install · How to audit a skill first