# Codex Premium Website Skills
[](https://github.com/TheGoat395/Codex-Skills/actions/workflows/validate.yml)
[](https://github.com/TheGoat395/Codex-Skills/releases/latest)
An unofficial, open-source, Codex-first Agent Skills library for premium frontend, website, design-engineering, and reliable agent workflows.
This repository is prepared around a curated `public-core` collection of 80 skills: design direction, frontend implementation, motion, accessibility, QA, content polish, maintainer safety, and agent-quality workflows.
## Works With
Built for Codex first. Portable to Agent Skills-compatible workflows and adaptable for Claude Code, Cursor, and Gemini CLI.
| Tool | Support Level | Path |
|---|---|---|
| Codex | Native supported | `scripts/install_skills.py` |
| Claude Code | Native-compatible `SKILL.md` folders | `adapters/claude-code/` |
| Cursor | Project rules adapter | `adapters/cursor/` |
| Gemini CLI | Project context adapter | `adapters/gemini-cli/` |
See `COMPATIBILITY.md` for support levels and setup notes.
## Quick Start
Get value in under 3 minutes:
```bash
git clone https://github.com/TheGoat395/Codex-Skills.git
cd Codex-Skills
python3 scripts/install_skills.py --dry-run
python3 scripts/install_skills.py
```
Start a new Codex conversation so skill discovery refreshes, then ask:
```text
Use the Codex website skills to review my homepage and tell me the highest-impact improvements before launch.
```
For a smaller first install:
```bash
python3 scripts/install_skills.py --collection taste-and-build-gates
```
For the newest reviewed skill changes, use the current `main` branch. Published release archives are versioned snapshots and may predate later fixes. To use a specific published snapshot, download it from [GitHub Releases](https://github.com/TheGoat395/Codex-Skills/releases/latest), extract it, then run the same install commands from the extracted folder.
## Premium Website Showcase
This repository includes a proof-of-output demo gallery:
```text
examples/premium-website-showcase/gallery/index.html
```
The showcase contains nine confirmed local websites and hero/portfolio demos generated through the premium website skill workflow. They are meant to show practical potential: install the skills, provide a prompt/reference/spec, and use the workflow to produce polished local frontend artifacts quickly.
See `DEMOS.md` for the complete list, `EVIDENCE.md` for the current evidence summary, and `INSTALL.md` for local install and preview instructions.
## Measured Benchmarks
Two paired baseline-vs-skill-assisted runs are checked in with fixed prompts, saved source, explicit limitations, and scoring against `BENCHMARKS.md`.
| Run | Baseline | Skill-assisted | Change | Evidence |
|---|---:|---:|---:|---|
| [Atlas](examples/benchmarks/measured-premium-website-template/runs/2026-07-10-atlas/) | 19/30 | 23/30 | +4 | Source and structural checks |
| [Signal / Matter](examples/benchmarks/measured-premium-website-template/runs/2026-07-10-signal-matter/) | 23/30 | 28/30 | +5 | Desktop/mobile screenshots and automated Chromium QA |
The Signal / Matter run used the same prompt, model, and three-file constraint in isolated directories. Both conditions were capable; the assisted workflow produced stronger visual differentiation, implementation detail, accessible state language, and first-viewport behavior. These are repository rubric scores, not an OpenAI evaluation, endorsement, or universal causal claim.
## Share Your Work
Used these skills on a real project? [Open a Showcase issue](https://github.com/TheGoat395/Codex-Skills/issues/new?template=showcase.yml) with the skills used, what changed, and any screenshots, links, or workflow notes that would help others evaluate the library.
For stronger evidence, submit a measured benchmark using `examples/benchmarks/measured-premium-website-template/`.
## What This Is
- A domain-specific Codex skill pack for modern website and app work.
- A Codex-first skill library with adapter paths for Claude Code, Cursor, and Gemini CLI workflows.
- A curated public core, not a loose dump of every local skill.
- A practical toolkit for React, Next.js, Tailwind, GSAP/Motion, QA, copy, accessibility, source research, visual polish, and evidence-led agent work.
- A repo designed to be inspected, installed selectively, benchmarked, and improved in public.
## What This Is Not
- Not an official OpenAI project.
- Not a guarantee that Codex will produce perfect design or production-ready code without review.
- Not a replacement for project context, real assets, rendered browser inspection, accessibility testing, or engineering judgment.
- Not a reason to install every skill at once. The default install is `public-core`.
- Not a security guarantee. Inspect the repo and installer before running anything.
## Repository Contents
- `skills/` contains individual Codex skill folders.
- `COMPATIBILITY.md` documents Codex, Claude Code, Cursor, and Gemini CLI usage paths.
- `CATALOG.md` gives a cleaner audience-facing map of collections and high-value skills.
- `adapters/` contains Claude Code, Cursor, and Gemini CLI adapter guidance.
- `curated_collections.json` defines the default 80-skill public core and smaller installable subcollections.
- `scripts/install_skills.py` installs `public-core` by default or selected collections with `--collection`.
- `THIRD_PARTY_NOTICES.md` records license and notice expectations for the public core.
- `EVIDENCE.md`, `BENCHMARKS.md`, `SKILL_QUALITY_STANDARD.md`, and `SECURITY.md` define how this repo should be evaluated and maintained.
- `manifest.json` and `SKILL_INVENTORY.md` are generated inventory files and should be regenerated before a public release.
- `examples/premium-website-showcase/` contains the local proof-of-output gallery and demo preview files.
- `DEMOS.md` documents the confirmed demo set and local preview paths.
- `INSTALL.md` gives a compact install and demo-preview flow.
- `MAINTAINERS.md`, `docs/maintainer-workflow.md`, and `docs/usage-signals.md` document release checks, evidence levels, and public usage signals.
## Install
Clone the repo, inspect it, then dry-run the default public core:
```bash
python3 scripts/install_skills.py --dry-run
```
Install the default 80-skill public core:
```bash
python3 scripts/install_skills.py
```
Install a focused collection:
```bash
python3 scripts/install_skills.py --collection motion-core --dry-run
python3 scripts/install_skills.py --collection motion-core
```
List available collections:
```bash
python3 scripts/install_skills.py --list-collections
```
Replace existing skills only after review:
```bash
python3 scripts/install_skills.py --replace
```
The installer uses `$CODEX_HOME/skills` when `CODEX_HOME` is set, otherwise `~/.codex/skills`; an explicit `--dest` takes precedence. It skips existing skills by default. When `--replace` is used, existing skill folders are moved into a timestamped backup folder inside the selected destination.
After installing, start a new Codex conversation so skill discovery refreshes.
## Public Core Collections
- `public-core`: the full 80-skill public release.
- `taste-and-build-gates`: design direction, build gates, anti-generic review, browser inspection, handoff.
- `visual-direction`: references, art direction, typography, color, layout, imagery.
- `frontend-implementation`: React, Next.js, Tailwind, primitives, states, forms, accessibility.
- `qa-production-core`: accessibility, performance, responsive, cross-browser, visual regression, SEO.
- `motion-core`: motion language, React motion, easing, jank prevention, reduced motion, GSAP performance.
- `content-conversion-core`: copy, IA, CTA language, conversion polish, proof.
- `maintainer-safety-core`: checkpoint, undo/revert, install-safety review.
- `agent-quality-core`: critical and logical thinking, creativity, research, evidence reconciliation, debugging, agent architecture, memory provenance, evaluation, and completion verification.
See `CATALOG.md` for the audience-facing catalog and `CURATED_COLLECTIONS.md` for the full map.
## Quality Bar
Every skill should have:
- a narrow trigger and clear description
- concrete actions rather than vague taste language
- explicit boundaries for when it should not run
- no hidden network, filesystem, or dependency side effects
- project-local dependency guidance
- verification steps when it changes code, frontend output, or agent workflow state
See `SKILL_QUALITY_STANDARD.md` for contribution criteria.
## Licensing
The root `LICENSE` covers original repository packaging, docs, scripts, and skills that do not carry separate notices. If a skill folder includes a more specific notice, that notice controls for that folder. See `THIRD_PARTY_NOTICES.md`.
## Benchmarking
The goal is not to claim that more skills automatically make better output. The goal is to show measurable workflow improvement. Benchmarks should compare the same task with and without selected skill collections, including screenshots, build logs, accessibility checks, browser QA notes, and final code links.
Use `BENCHMARKS.md`, `examples/README.md`, and `examples/benchmarks/measured-premium-website-template/` as the reporting protocol. Atlas is the first source-checked comparison; Signal / Matter adds rendered desktop/mobile evidence and automated Chromium QA. Future runs should preserve the same controls and disclose operator corrections and browser limitations.
## Open Source Positioning
This project is best described as:
> An open-source Codex skill library for premium frontend, website, design-engineering, and evidence-led agent workflows.
The strongest public contribution is not the size of the library. It is the curation: an installable public core, focused groups, quality standards, safety policy, benchmarks, examples, and real maintenance practices for people building with Codex.
查看 README 源文件