skill
agent-skills
Open-source skills for AI coding agents: product-frontend (everyday product UI and the code behind it) and job-search.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/codad5/agent-skills itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 22, 2026Agent skills
Open-source skills for AI coding agents: Claude Code, Cursor, Codex, Gemini CLI, Copilot and any other agent that reads SKILL.md. Each skill is a folder of plain Markdown: a short SKILL.md that routes the agent through phases, plus one file per phase.
Skills
| Skill | For | Install |
|---|---|---|
product-frontend | Everyday product UI and the code behind it: dashboards, admin tools, chat apps, forms, marketing pages | npx skills add codad5/agent-skills --skill product-frontend |
job-search | Finding real job listings and preparing every application (it never applies for you) | npx skills add codad5/agent-skills --skill job-search |
social-optimization | Auditing and improving your professional profiles: LinkedIn, GitHub, X, Instagram, portfolio site | npx skills add codad5/agent-skills --skill social-optimization |
See every skill in the repo, with descriptions, without installing anything:
npx skills add codad5/agent-skills --list
Install them all at once with npx skills add codad5/agent-skills.
Before you start
- Node.js 22 or newer, for
npx. That's all: no account, no login, no config. The repo is public. - Run the command inside the folder you want the skill in: your project for
product-frontend, or your job-search working folder forjob-search, or your portfolio or notes folder forsocial-optimization. - Useful flags:
-a claude-codeinstalls for one agent and skips the agent prompt (other values includecursor,codex,gemini-cli,github-copilot).-ginstalls for every project on this machine instead of just this folder.-yaccepts the defaults without prompting.
- After installing, start a new agent session so it picks the skill up.
- Check what's installed with
npx skills list.
Or let your agent set it up
Copy this into your agent as it is: nothing to fill in. Add a skill name at the end if you already know which one you want; leave it off and the agent will help you choose.
Set up a skill for me from https://github.com/codad5/agent-skills.
1. Run `npx skills add codad5/agent-skills --list` to see every skill in the repo with
its description. If npx isn't available, read the repo's README and each
skills/*/SKILL.md instead.
2. If I named a skill below, use it. If I didn't, show me the list in one line per
skill, ask what I want to get done, and recommend the skill that fits best with
one line on why. Wait for my answer.
3. Work out which agent you are and install the chosen skill for this folder only:
`npx skills add codad5/agent-skills --skill <chosen skill> -a <your agent id> -y`
(claude-code, cursor, codex, gemini-cli, github-copilot, windsurf or opencode).
Without npx, clone the repo and copy skills/<chosen skill>/ into this folder's
skills directory (.claude/skills/ for Claude Code, .agents/skills/ for most others).
4. Read the installed SKILL.md and tell me in two lines what it does, what it needs
from me (tools, files, permissions) and whether I should start a new session.
Skill (optional):
product-frontend
Plans and builds the everyday parts of a product's frontend: both how it looks and how it works. It reads your codebase first and works out whether the real problem is the UI, the code underneath (rendering, data fetching, state, types, performance) or both. It recommends working only where it pays, and it obeys "keep the UI as it is" or "only restyle".
Not for showpiece work: games, 3D, heavy animation, illustration or experimental interactive art.
Set up
cd your-project
npx skills add codad5/agent-skills --skill product-frontend -a claude-code
Helps if available, but optional:
- a browser tool for the agent (Claude Code's built-in browser, a Playwright MCP server, or the Claude in Chrome extension), so it can see and screenshot your screens. Without one it asks you for screenshots.
- permission to run your dev server. It won't start one unless you agree, and it uses a non-default port.
What to do next
- Open a new agent session in the project.
- Ask in your own words, or type
/product-frontendin Claude Code. For example:- "Revamp our dashboard, it looks generated."
- "Keep the UI exactly as it is, but our pages are slow and the data fetching is a mess."
- "Set up a design system before we build the settings pages."
- "Polish this page without moving anything."
- Expect, in order:
- a short read of your code, with the UI and the code each scored out of 10;
- one round of questions with a recommended answer for each: what the project is to you (MVP, startup, client work, side project), UI or code or both, and scope;
- effort options sized in days, commits and tokens, with a recommendation and whether it can go in gradually or needs a rebuild;
- visual directions (UI work) or architecture options (code work) to choose from;
- a phased plan for you to approve. Nothing is built before you say yes.
- Then it builds phase by phase. After each phase it sends a short update and stops at the review points in the plan.
Plans, decisions and progress are saved in your project's docs (docs/design/ by default), so a later session can pick up where the last one stopped.
Files
| File | What it covers |
|---|---|
| SKILL.md | The order of phases and the hard rules |
| 01-context-scan.md | Reading the code, docs, history and screens; scoring |
| 02-intake.md | Which questions to ask and how |
| 03-sizing-and-approach.md | Project type, effort levels, patch vs rewrite |
| 04-directions-and-proposals.md | Showing visual options; colour ramps; decoding screenshots |
| 05-architecture-and-data.md | Rendering, the data seam, sample data, state, performance |
| 06-design-system.md | Tokens, surfaces, components, enforcement |
| 07-plan-and-build.md | The phased plan, the per-screen loop, tracking |
| 08-craft-checklist.md | The final quality pass |
| 09-layout-rules.md | Optional box layout rules (gap-only spacing, tokens only) |
| 10-working-agreements.md | Dev server, tests, commits, updates |
| 11-related-skills.md | Finding other skills that fit your stack |
| 12-reference-cases.md | The real projects behind the rules |
| 13-ui-styles.md | UI styles, what each suits, and where to find references |
| 14-motion-and-feedback.md | Instant feedback, optimistic updates, loading, slow and offline networks, motion |
job-search
Searches live job boards and company career pages, filters listings against your real constraints (location, visa, remote vs on-site pay floor, role fit) and gives you a shortlist to prune. Then, for each job you approve, it writes one complete doc: how to apply, every form question answered, a tailored resume and cover letter, an ATS keyword check, when to apply, and an honest fit score.
It never applies, sends emails or submits forms. You do that yourself.
Set up
mkdir job-search && cd job-search # any folder you'll keep your search in
npx skills add codad5/agent-skills --skill job-search -a claude-code
Needed:
- a browser tool for the agent (Claude Code's built-in browser, a Playwright MCP server, or the Claude in Chrome extension). Listings must be read live, so without one the skill stops and asks you to enable one.
Recommended:
- your current resume or CV in the folder (Markdown, PDF or DOCX), so it can match against it and tailor copies;
- being logged in to LinkedIn or other job boards in that browser. It can work without this, but results are better with it. It never changes your account settings.
What to do next
- Open a new agent session in the folder.
- Ask: "Find me jobs", or type
/job-searchin Claude Code. - Answer one round of questions:
- target roles and seniority;
- where you can work remotely without a visa, where you'd relocate with sponsorship, and which cities you'd accept on-site;
- your pay floor;
- how many jobs and how recent;
- which platforms and companies.
- Review the shortlist (
YYYY-MM-DD/shortlist.md) and say which jobs to keep. - Get one doc per approved job in the same date folder, and a
00-INDEX.mdlisting every search, its jobs, their status and what the market keeps asking for that your resume doesn't show. - Apply yourself, using the docs.
Files
| File | What it covers |
|---|---|
| SKILL.md | The order of phases, output layout, hard rules |
| 01-intake.md | The questions it asks before searching |
| 02-shortlist.md | Searching, filtering, the shortlist |
| 03-per-job-doc.md | The per-job document template |
| 04-resume-and-ats.md | Tailoring resumes and ATS-safe PDFs |
social-optimization
Audits and improves a person's professional presence: LinkedIn, GitHub, X, Instagram and their portfolio site. It researches who the person actually is before writing a word and builds a private dossier: their own words, their stated preferences, and an inconsistency table across platforms. It drafts every change for approval, then works through one platform at a time, re-checking each save after a fresh reload. Banners are built as HTML and rendered, never AI-generated.
It asks before visiting your logged-in profiles, before asking your own AI assistants about you, and before every live edit. It never invents numbers, never takes credit that belongs to someone else, and never posts new content or messages anyone.
Set up
cd your-portfolio-or-notes-folder
npx skills add codad5/agent-skills --skill social-optimization -a claude-code
Needed:
- a browser tool for the agent (Claude Code's built-in browser, a Playwright MCP server, or the Claude in Chrome extension), ideally with you logged in to the platforms in that browser.
Recommended:
- your resume or CV and any existing bio text in the folder;
.tmp/in your.gitignore. The private dossier and drafts live there and must never be committed.
What to do next
- Open a new agent session in the folder.
- Ask: "Optimise my LinkedIn", "Audit my personal brand", or type
/social-optimizationin Claude Code. - Approve the research steps it asks about, then answer its questions about how you want to be described (your own label, who drove which results, what to feature).
- Review the drafted copy before anything goes live.
- Approve each platform pass. It works through them in order: LinkedIn, GitHub, X, Instagram, then your site.
- Get a final checklist: what's done and re-verified, what you need to do by hand and why, and what it recommends but didn't apply.
Its dossier works with job-search too: that skill reads it instead of asking the same questions again.
Files
| File | What it covers |
|---|---|
| SKILL.md | Order of phases, hard rules, where files go |
| 01-research-identity.md | Researching the person and writing the dossier |
| 02-linkedin.md | LinkedIn: order of work and known gotchas |
| 03-github.md | GitHub profile, pinned repos, README widgets |
| 04-x-twitter.md | X: name, bio, photo and banner, tone |
| 05-instagram.md | Instagram, as optional polish |
| 06-portfolio-site.md | Keeping the site's story consistent with the profiles |
| 07-banner-generation.md | Building banners as HTML and rendering them |
Install without npx
git clone https://github.com/codad5/agent-skills.git
Copy the skill's folder to where your agent looks for skills:
| Agent | This project | Every project |
|---|---|---|
| Claude Code | .claude/skills/<skill-name>/ | ~/.claude/skills/<skill-name>/ |
Agents that read .agents/ (Codex, Cursor and others) | .agents/skills/<skill-name>/ | ~/.agents/skills/<skill-name>/ |
mkdir -p .claude/skills
cp -r agent-skills/skills/product-frontend .claude/skills/
On Windows PowerShell: Copy-Item -Recurse agent-skills\skills\product-frontend .claude\skills\.
To update later, run the npx skills add command again, or pull the repo and copy the folder again.
Find more skills
npx skills find design system
npx skills find nextjs
Or browse skills.sh. product-frontend also looks for skills that fit your stack and suggests them, but it asks before installing anything.
Contributing
Issues and pull requests are welcome. Keep the style: plain imperative prose, every rule with a reason, a "What NOT to do" section at the end of each phase file, and nothing personal or project-specific baked in.