46 Practical Codex Skills from Composio Community

Codex skills are how you teach OpenAI's Codex CLI and API to handle the same task the same way, every time. The composio-community has published 46 of them under ComposioHQ/awesome-codex-skills — modular instruction bundles for development, productivity, communication, and data work — installable with a single command from the repo's bundled skill-installer.

Why This Skill Matters

The collection is small enough to read in an afternoon and broad enough to cover the workflow chunks you actually repeat. Categories:

• Development & Code Tools — code review, refactoring patterns, test scaffolds.
• Productivity & Collaboration — meeting notes, action items, status updates.
• Communication & Writing — drafting, summarizing, tone adjustment.
• Data & Analysis — CSV inspection, chart descriptions, data sanity checks.
• Meta & Utilities — skill-of-skills glue.

What's notable is that each skill is small, scoped, and standalone. You can install one, two, or all forty-six without dragging in unrelated dependencies. Skills are triggered by natural-language matching rather than explicit invocation, so naming and description matter — pick skills whose descriptions match how you actually talk to Codex.

Installation

Install via the bundled skill-installer:

python skill-installer/scripts/install-skill-from-github.py \
  --repo ComposioHQ/awesome-codex-skills \
  --path meeting-notes-and-actions

Replace --path with the skill you want (one at a time). The README's quickstart uses ComposioHQ/awesome-codex-skills as the repo; the actual GitHub organization is composio-community. If the install script doesn't find the repo, swap the org in the --repo flag.

After installation, restart Codex so it reloads metadata. The skills won't be discoverable until you do.

Real Workflow: Turn Meeting Transcripts into Action Items

A realistic scenario, not "hello world": after every team meeting, you need a summary, decisions list, and owner-tagged action items. The meeting-notes-and-actions skill is built for exactly this.

The workflow:

1. Install the skill (above).

2. Paste your meeting transcript into Codex.

3. Prompt:

   Use the meeting-notes-and-actions skill. From the transcript below,
   produce:
     - a 5-bullet executive summary
     - a decisions list
     - an action-items list with owners and due dates
   
   <paste transcript>
   

Expected output: three structured sections. Example:

Summary:
- ...

Decisions:
- D1: Use Postgres over SQLite (owner: @alice, 2026-08-19)

Action items:
- A1: Draft schema migration (owner: @bob, due 2026-08-22)
- A2: Set up staging DB (owner: @carol, due 2026-08-20)

Owners and due dates come from the transcript — if the transcript doesn't name them, the skill leaves placeholders, which is the right behavior.

Real Workflow: Data Sanity Checks on a New Dataset

Another realistic scenario: you receive a CSV you haven't seen before, and you want Codex to flag what's odd before you start real analysis.

The workflow:

1. Install the relevant data-analysis skill.

2. Load the CSV:

   Use the data-sanity-check skill on this CSV. Report: row count, column
   types, missing-value counts, outliers (3σ), duplicate rows, and any
   values that look malformed (negative ages, future timestamps, etc.).
   
   <paste CSV path or first 100 rows>
   

Expected output: a structured report with row count, per-column missingness, a list of outliers, and a list of suspected-malformed values. It's not a substitute for full data validation, but it catches the obvious things in seconds.

Tips

• Install one skill at a time. The installer takes a --path, not a list. Install, try, then add the next.
• Restart Codex after each install. Skills are matched by natural language against the loaded skill metadata. Until you restart, new skills aren't visible.
• Read each skill's description. Skills trigger on natural-language match — a description that doesn't match how you prompt will never fire.
• Mind the third-party deps. Some skills (connect, helium-mcp, langsmith-fetch) need external service credentials. Skip those if you don't have the accounts.

When Not to Use This

• You're not on Codex. These skills are scoped to the Codex CLI and API. Other agents (Claude Code, Cursor, Gemini CLI) won't load them.
• You need an explicit-invoke pattern. Skills trigger on natural-language match, not on explicit commands. If you want a hard "run this exact skill" workflow, layer that on top.
• You can't restart Codex between installs. Skills only become visible after a restart. In an automation pipeline, batch installs upfront rather than rolling them in.

See the leaderboard for more skills.