REPO PACKAGE
openai skills
OFFICIAL43skills
9ready to use
LABSCO SUMMARY
Eight of the 43 are Figma-to-code skills (figma, figma-use, and six companions for design-system rules, new files, and full-screen generation from a codebase), four are one-per-platform deploy skills (cloudflare-deploy, netlify-deploy, render-deploy, vercel-deploy), and four cover Notion end to end (capture, meeting-intelligence, research-documentation, spec-to-implementation). Two handle GitHub PR review and CI triage (gh-address-comments, gh-fix-ci), and the rest are standalone: PDF handling, Jupyter notebooks, Playwright browser automation, screenshots, image generation, text-to-speech, transcription, and three meta-skills for building and installing further skills (skill-creator, skill-installer, plugin-creator).
This is Codex's own catalogue, so most of it assumes you are running Codex rather than a generic agent — several skills reference Codex-specific mechanics like $imagegen composition or the .codex-plugin/plugin.json layout that a different client will not recognise. Eight of the 43 need nothing beyond installing the skill; the other 35 split three ways — 12 need an account and credentials before they do anything, 12 need a local binary already on your machine, and 11 need an MCP server you have set up yourself. Check the label on each skill below before picking one; "install and go" is the minority case here.
READ THE FULL ANALYSIS
Why the deprecation matters for adoption. The README's own banner points new work at openai/plugins for current examples and at OpenAI's "Build plugins" guide for anyone adding a skill-only plugin. Nothing here has been removed, but new work is not meant to start in this repository.
Thirteen have already gone dead. Beyond the 43 we could verify as live, 13 more that were once catalogued for this repository have since disappeared, out of 56 total — consistent with a repository whose own maintainers say to look elsewhere for anything new.
Checked 19 September 2026 against what is currently live; the deprecation notice is openai's own, unedited.
WHAT'S INSIDE
43 showing · 43 totalReady to use
Nothing else to set up — install it and go.
43 of 43
SETUP
Showing all 43 skills
aspnet-core
Builds and reworks ASP.NET Core web apps the way Microsoft's current documentation says to, rather than the older patterns that still circulate online.
chatgpt-apps
Builds an app that runs inside ChatGPT — the server that does the work, plus the small interactive panel that appears in the conversation — following OpenAI's current published instructions.
cli-creator
Turns an API's documentation, a saved curl command or a script you keep re-running into a proper command you can type from anywhere on your machine.
cloudflare-deploy
Cloudflare sells dozens of overlapping products and half the work is knowing which of them you need; this picks them for what you are building and gets the thing running there.
define-goal
Pins down the goal before the work starts — what exactly will be true at the end, what number proves it, and when to stop and ask.
figma
Sets out how to pull a design out of Figma properly — the layer's data and a picture of it first, and Figma's own images rather than a new icon pack — before any code gets written from it.
figma-code-connect-components
Links the pieces of a Figma design to the code that actually builds them, so a designer and a developer end up looking at the same thing.
figma-create-design-system-rules
Writes a project's house rules — where components live, how they are named, which colours and spacing are allowed — into a file the coding assistant reads before building anything from a design.
figma-create-new-file
Makes a fresh, empty Figma design file or FigJam whiteboard and hands back the link to open it.
figma-generate-design
Builds a whole screen inside Figma from an existing page of an app, using the team's real buttons, colours and text styles rather than drawing look-alike boxes.
figma-generate-library
Builds the shared kit a design team works from — colours, text styles, reusable components — inside Figma so it matches what the code already uses, in stages with your approval between them.
figma-implement-design
Turns one specific frame of a Figma file into working code for your app, then checks the result against the design until the two match.
figma-use
The traps to avoid when a coding assistant edits a Figma file directly by script — the ones that otherwise fail quietly or leave half-built shapes on the canvas.
gh-address-comments
Collects the feedback people left on your pull request, says what each one would take to fix, and then fixes the ones you pick.
gh-fix-ci
Works out why the automated checks on a pull request are failing, pulls out the part of the log that explains it, and fixes it once you approve the plan.
hatch-pet
Draws a small animated mascot — idle, walking, sleeping and the rest of its moods — from a description, a company's branding or a reference picture, and packages it up ready for Codex to show.
imagegen
Makes a picture from a written description, or changes one you already have — swapping a background, removing an object, cutting a subject out.
jupyter-notebook
Starts a Jupyter notebook from a ready-made template — an experiment write-up or a teaching walkthrough — so it still runs top to bottom for whoever opens it next.
linear
Reads and changes the tickets in Linear from the chat — triaging bugs, planning a sprint, chasing blocked work — without opening Linear itself.
migrate-to-codex
Moves a project's Claude Code setup — its instructions, skills, agents and connected tools — into the files Codex reads instead, then reports what moved and what still needs a human.
netlify-deploy
Puts a website live on Netlify from the command line — matched to the right site first, then out to a preview link to check before the public one goes out.
notion-knowledge-capture
Takes what was said in a conversation and files it in Notion as something the team can find later: a how-to, an FAQ entry, or a record of a decision and the reasoning behind it.
notion-meeting-intelligence
Gets a meeting ready — an agenda or pre-read drafted in Notion from what the team has already written down, with an owner and a time slot against each item.
notion-research-documentation
Reads across the pages already sitting in Notion and writes up what they amount to — a short brief, a side-by-side comparison, or a full report — with a link back to every source.
notion-spec-to-implementation
Turns a written spec in Notion into a plan and a task list, each task small enough for a day or two, all linked back to the spec and kept current as the work moves.
openai-docs
Questions about OpenAI's products — including Codex, its coding agent — answered out of the current official documentation, fetched at the moment you ask rather than recalled from memory, with a link to where each answer came from.
Reads a PDF, or builds a new one, and looks at every page as a picture before handing it over — so clipped text and broken tables do not slip through.
playwright
Drives a real web browser from the terminal — clicking through pages, grabbing screenshots, pulling data off a page — without writing a test file for any of it.
playwright-interactive
Keeps one browser window open while a local app is being fixed, so each change can be reloaded and checked in place instead of restarting everything each time.
plugin-creator
Creates the folder and starter files for a new Codex plugin, with the blanks marked so all that is left is filling them in.
render-deploy
Works out what a project needs in order to run on Render — the web service, the database, the background worker — and puts it live there.
screenshot
Takes a picture of what is on the screen — the whole display, one app's window, or a rectangle you specify — and saves it to a file.
security-best-practices
Brings in the security rules that actually apply to the language and framework a project uses, so new code follows them from the start and the risky patterns already sitting in the codebase get pointed out instead of left alone.
security-ownership-map
Works out who actually knows each part of a codebase from its commit history, and flags the sensitive files only one person has ever touched.
security-threat-model
Goes through a repository asking how an attacker would come at it — what is worth taking, which way in, what would stop them — and writes the answer up, pointing at the code it is based on.
sentry
Pulls up the errors a live application is throwing, from Sentry, so you can see what is breaking without leaving the terminal.
skill-creator
Walks through building a new skill for Codex — what belongs in it, how to word the line that makes Codex reach for it, and which files go in the folder.
skill-installer
Installs a ready-made skill into Codex, either from OpenAI's own curated list or from any GitHub repository you point it at.
speech
Reads written text aloud in one of a set of built-in voices and saves it as an audio file — a single clip, or a whole batch in one run.
transcribe
Turns a recording into written text, and can mark who said which part when several people are talking.
vercel-deploy
Gets a project onto Vercel in one step, deploying to a private preview link unless you specifically ask for the live site.
winui-app
Sets up a Windows desktop app from scratch and does not call it done until the window has actually opened on the screen.
yeet
The whole end-of-work ritual in one go — the changes go in, the branch goes up, and the pull request is sitting there waiting for review.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (43 skills)
npx skills add openai/skillsSINGLE SKILLOr install a single skill
npx skills add openai/skills --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.