Claude Code skills
Skills
9 skills I use to do product engineering with an agent: plan the work, review it, keep it patched, ship it, and then get people to find it. Each one is a single markdown file, and each ships as a slash command.
Install
Claude Code plugin
All nine skills in one plugin. Run these inside Claude Code, then /reload-plugins.
/plugin marketplace add joshghent/skills
/plugin install joshghent-skills@joshghent-skills
Skills CLI
Pick one skill and install it into Claude Code, Cursor, Codex, or any agent the skills CLI supports.
npx skills add joshghent/skills --skill groundwork
Keep them current with /plugin marketplace update joshghent-skills.
Or skip the tooling: copy any skill folder into ~/.claude/skills/.
Everything is on GitHub.
What is a skill?
A skill is a markdown file of instructions that an agent loads when a request matches it. It is the difference between asking for a code review and handing over the checklist you would use yourself.
The problem
Ask an agent to "update the dependencies" and you get a different process each time. It forgets the build, skips the tests, or bumps a major version in the same PR as a security fix. You end up re-explaining how you want it done, every time.
The fix
Write the process down once. A skill sets out the steps, what to check, what never to do, and what the output should look like. The agent follows it the same way on every repo, and you can trigger it with one slash command.
Why it compounds
When a run goes wrong, you fix the skill rather than the prompt. That lesson then applies to every future run, on every project. Over a few months the skill holds more of your judgement than you could type out on any single day.
The skills
Plan and set up
Before the first line of code: turn the plan into a timeline, and make sure anyone can run the repo from a fresh clone.
-
gantarr
/gantarr
Turns a plan, roadmap, or the conversation so far into a valid Gantarr JSON file you can import to render a Gantt chart.
-
groundwork
New/groundwork
Walks the fresh-clone path for real, scores local dev readiness, fixes the gaps, and opens one PR with a before/after scorecard.
Review
A second opinion on the code and the interface, ranked by what is most likely to hurt you in production or cost you a conversion.
-
sentinel
/sentinel
Audits the whole repo for quality, test coverage, CI, security, and AI-agent fragility, ranked by production risk.
-
design-review
/design-review
Reviews a UI for conversion, UX, accessibility, and performance against Apple, FT, and OpenAI standards, and calls out AI design slop.
Maintain and ship
The chores that pile up: patching dependencies, clearing the PR queue, and checking the deploy actually went out.
-
warden
/warden
Pulls production errors, then patches security alerts and bumps dependencies, verified by your own build and tests, in one PR.
-
conductor
/conductor
Merges every ready PR newest first, each one green on CI, then confirms the deploy shipped and the live app answers.
Grow
After it ships: write about it, get it found in search and AI answers, and run a weekly loop on the numbers.
-
blogger
/blogger
Turns shipped commits into dated blog posts, changelog entries, and social snippets written in your site's voice.
-
beacon
/beacon
Reviews SEO and AI-answer visibility from Search Console, Ahrefs, and PostHog data, then fixes pages, metadata, and schema.
-
growth-review
/growth-review
Runs a weekly growth loop: health checks, this week against last, a dated log entry, then ships the one highest-leverage change.
What’s changed
Full history →-
- plugin Every skill can now be installed on its own with the skills CLI, into any agent it supports.
-
v1.2.0
- groundwork New Walks the fresh-clone path, scores ten setup steps, fixes the gaps, and opens one PR with the before/after scorecard.
- groundwork Also checks test coverage (is there a command, does CI enforce it, which critical paths are untested) and whether deploys are automated and gated on CI.
-
v1.1.0
- warden Added /warden security: collects every open high and critical Dependabot, code scanning, and secret scanning alert across a repo, a list of repos, or an org, and opens one PR that fixes or triages them.
- conductor Merges security-sweep PRs first, then re-counts open high and critical alerts after the deploy.
- plugin Releases are now versioned with changesets, so /plugin marketplace update pulls new skills reliably.
-
- growth-review New The weekly loop: check the site is actually up and tracking, log the week's numbers, then build one improvement.
- plugin The plugin and marketplace are now both called joshghent-skills.
-
- beacon New SEO and GEO review driven by real Search Console, Ahrefs, and PostHog data, including whether AI crawlers are allowed in.
- conductor New A merge train for ready PRs that never force-merges, then watches the deploy and smoke-checks the live app.
- warden Prioritises the dependency sweep by live production errors from Sentry, New Relic, Cloudflare Workers, and others.
-
- plugin Fixed the plugin install so Claude Code discovers every skill.
-
- gantarr New Plan in, importable Gantt timeline out, validated before it is handed over.
- plugin All skills bundled into a single plugin, so install is two commands at most.
-
- design-review New A senior product designer's critique of a rendered UI, across viewports and keyboard states.
- warden New Dependency and security maintenance, including end-of-life runtime upgrades to the current LTS.
- plugin The repo became a Claude Code plugin marketplace.
-