The Skill Layer
what's a skill
A skill is a reusable slash command for Claude Code. It's a markdown file with structured instructions that Claude executes like a playbook. Instead of typing 'check my site for accessibility issues against WCAG guidelines,' you type /a11y and the skill handles scope, methodology, output format, and follow-up. No code, no plugins, no SDK. If you can write a checklist, you can write a skill.
The difference isn't convenience. It's consistency. Every session runs the same way regardless of how much context Claude has lost since the last one.
what 20 skills looks like
operations
| Skill | What it does |
|---|---|
/onboard | Digests a repo's docs, session logs, architecture, and git state so Claude is productive in under a minute |
/session | Generates an end-of-session log: what changed, decisions made, known issues, next steps |
/ship | Pre-deploy safety checklist |
/report | Full health report on any project or all projects at once |
audits
| Skill | What it does |
|---|---|
/a11y | WCAG accessibility audit against source code and optionally live URLs |
/seo | SEO health audit with scoring and gap identification |
/perf | Bundle size, image weight, render-blocking resources, Core Web Vitals |
/privacy | Data collection, consent flows, exposed secrets, privacy policy accuracy |
/vuln | Vulnerability scan across dependencies and code patterns |
monitoring
| Skill | What it does |
|---|---|
/pulse | Uptime, SSL, response time, and basic health across all live sites |
/sentry | Errors, issues, and logs across all projects |
/metrics | Analytics report using PostHog custom events and traffic data |
/cost | Monthly burn rate across all services, flag tier limits, compare against revenue |
content
| Skill | What it does |
|---|---|
/blog | Scaffold a blog post with domain-specific verification built in |
/social | Extract social media posts from a blog post |
/voice | Check content against brand voice rules |
/verify | Verify factual claims in content against authoritative sources |
project-specific
| Skill | What it does |
|---|---|
/transit | Look up planetary positions for any date using Swiss Ephemeris |
/update | Run a daily content update workflow |
/solo-stack-update | Research pricing/tier changes and update structured data |
why skills matter
Every skill encodes a decision that was made once and shouldn't need to be made again.
/verify exists because Claude hallucinated a date in an early post. /ship exists because a deploy went out with a broken API route. /session exists because without structured handoff notes, the next conversation starts from zero. Each one is a scar turned into a guardrail.
Skills also compound. /onboard reads the session logs that /session wrote. /blog calls /verify before publishing. /report orchestrates /pulse, /sentry, and /metrics into a single health check. The output of one session becomes the context for the next.
how to start
Pick the thing you explain to Claude most often. The workflow you repeat, the constraints you re-teach, the checklist you run manually. Write it as a markdown file in .claude/skills/your-skill-name/SKILL.md with clear instructions. That's your first skill.
From there, build the two that changed everything for me: /onboard (so Claude knows what it's working on) and /session (so the next conversation knows what happened in this one). Those two alone solve the biggest problem with AI-assisted development, which is context loss between sessions.
The tool is capable enough. The bottleneck is how much of your operational knowledge lives in your head versus in a format Claude can execute.