OpenAI Codex
Browse 35 Agent Skills recorded as compatible with OpenAI Codex.
Compatible skills
A fast Rust browser-automation CLI for AI agents using Chrome via CDP with accessibility-tree snapshots; a discovery stub that loads live workflows from the CLI itself.
Generate up to 25 diverse ad creative hypotheses — an angle × format matrix with headlines, visual prompts, and a testing rationale for each.
Converges on one visual direction by proposing 3-6 named, structurally distinct directions with rationale and reusable prompt stubs, plus one reference image each.
Builds 2-4 named visual territories with palette, texture, typography, and 4-6 generated tiles each so a team can align and sign off on a direction before production spend.
Turns a concrete offer (price, bundle, guarantee, deadline) into 3-5 hero promo visuals that make the value legible within two seconds, never inventing prices or terms.
Takes a chosen visual to production-ready state via diagnose-then-fix passes: inpaint artifacts, balance composition, color-grade, sharpen text, and conform to final size.
Maps business positioning (category, differentiator, audience, foil) onto a defensible visual language spec where every color, type, and imagery rule traces to a strategic reason.
Orchestrator that reads a broad creative goal and routes to the right creative-* sub-skill, sequencing direction, then assets, then polish for a coherent campaign set.
Generates believable product-in-use lifestyle scenes from concrete usage moments (who/where/when/doing-what) with natural light, keeping the product the clear hero.
Produces a consistent multi-angle ecommerce shot set of one product (hero, front, side, macro, packaging) with identity, lighting, and surface locked across every shot.
Deploys apps to Vercel, defaulting to preview, by detecting project/git/CLI state and choosing git-push, vercel deploy, link-first, or a no-auth sandbox fallback.
Audits an existing UI (URL, screenshot, or code) against usability, hierarchy, accessibility, and conversion heuristics, returning a severity-ranked issue list with fixes.
Clarifies the brief before design work by interrogating goal, user, JTBD, constraints, success, and scope, then writing a context.md the other design skills consume.
Generates 2-4 structurally distinct UI directions for a screen or flow, each with layout logic, component choices, and explicit optimizes/sacrifices mapped to the success criteria.
Reproduces a reference mockup as a runnable frontend, extracting its design system then screenshot-comparing the build and iterating until the gap is minor and explained.
Wires static screens into a clickable runnable prototype with routing, interaction states, and stubbed data, then verifies the whole happy path plus an error state in a browser.
Conformance-checks a built prototype against its mockup, context, and success criteria, flagging visual, behavioral, and a11y drift with expected-vs-actual and a verdict.
Gathers outside UX signal via convention scans, 3-5 competitor reviews, and pain mining from forums, synthesizing pains/patterns/gaps into a cited research.md for ideation.
Prepares a QA'd prototype to run cleanly and returns a working link plus run instructions and a review note, choosing local serve, tunnel, or deploy only with explicit consent.
Rebuilds a live site as a clean editable local frontend by extracting design tokens into url-dna.md, then reconstructing in your stack and screenshot-comparing it.
A portable, framework-agnostic React/React Native architecture style: feature modules with page directories, barrel-only imports, server-vs-UI state split, and component-promotion rules.
A framework-agnostic discipline for type safety at the network edge: one typed API client, parse-don't-validate, a single response envelope, one ApiError type, and branded IDs.
A portable Lighthouse CI performance gate that blocks PRs unless the production build meets Core Web Vitals budgets and category score floors, with median-of-N runs.
A framework-agnostic field-side observability system: one typed event taxonomy, best-effort non-blocking provider fan-out, real-user Web Vitals, error reporting, and consent gating.
A framework-agnostic write-path discipline: the cancel-snapshot-patch-rollback-invalidate optimistic lifecycle, once-per-intent idempotency keys, and multi-cache coherence.
A portable SEO system for React/React-Native-web: one metadata source, canonical URLs, sitemap, robots, RSS, and typed JSON-LD structured data.
Orchestrator for the design-* set that frames the brief, explores options, builds the chosen direction, gates with QA, and ships, routing each request to a sub-skill.
Deploys and manages Vercel projects using token-based auth read from the environment, covering deploy, env vars, domains, and inspection without interactive login.
React composition patterns that scale—compound components, lifted state, context interfaces, explicit variants over boolean props—plus React 19 API changes.
Runs an observability-first Vercel cost and performance audit, collecting production metrics before investigating metric-backed candidates and producing ranked, file-grounded recommendations.
Vercel Engineering's 70 React/Next.js performance rules across 8 priority categories—waterfalls, bundle size, server-side, re-renders—for writing, reviewing, and refactoring code.
React Native and Expo best practices across list performance, Reanimated animations, navigation, UI patterns, and monorepo config, organized by priority and impact.
Implements native-feeling animations with React's View Transition API—shared elements, Suspense reveals, list reorder, directional navigation—without third-party animation libraries.
Review UI code against Vercel's Web Interface Guidelines — accessibility, UX, and design best practices — with terse file:line findings.
Reviews docs and prose against Vercel's Writing Guidelines by fetching the latest rules and reporting findings in a terse file:line format.