Claude Code Skills

I don’t know what I don’t know. So I built skills that do.

---

Why this exists

I’m an engineer building a product solo. I can write code all day — but that’s maybe 30% of what shipping a product actually takes. UX design, product strategy, questioning my own decisions, catching what AI missed — I don’t have a team of specialists for any of it.

So I started giving Claude actual knowledge bases instead of just asking it to “be an expert.” Audit frameworks, questioning methodologies, design principles — the stuff specialists carry in their heads after years of work. Now every skill cites its reasoning, teaches me something, and produces output I can actually use.

These skills are tested on a live product. They’ve caught critical issues I would’ve shipped blind.

---

Install

Each skill is self-contained. Grab what you need.

# Pick a skill and install it
npx degit notmanas/claude-code-skills/skills/ .claude/skills/

# Example: install the UX expert
npx degit notmanas/claude-code-skills/skills/ux-expert .claude/skills/ux-expert

# Example: install devil's advocate
npx degit notmanas/claude-code-skills/skills/devils-advocate .claude/skills/devils-advocate

No config. No dependencies. Just markdown files that Claude Code picks up automatically.

---

Available Skills

ux-expert — UX audits backed by real methodology

Audits your pages across 8 dimensions (hierarchy, cognitive load, screen real estate, interaction cost…). Every finding cites the UX principle being violated. Produces implementation-ready redesign specs with wireframes.

/ux-expert audit the dashboard

Knowledge base: ~2,000 lines covering Gestalt principles, Shneiderman’s mantra, cognitive load theory, component library decision guides.

Read more

---

devils-advocate — “Are you sure about that?”

Challenges AI-generated plans, code, and decisions before you commit. Uses pre-mortem analysis, inversion thinking, and Socratic questioning to find what was missed — hidden assumptions, failure modes, and the optimistic shortcuts AI never questions.

/devils-advocate review the auth flow

Pairs with any other skill as a review layer:

/ux-expert  →  design the page   →  /devils-advocate  →  challenge the design
Claude Code →  build the feature  →  /devils-advocate  →  find what was missed

Knowledge base: ~1,100 lines covering questioning frameworks, 11 engineering blind spot categories, and AI-specific failure patterns.

Read more

---

What makes these different from “just prompting”

They teach you while they work. Every finding comes with the reasoning behind it. You’re not just getting a review — you’re building intuition you didn’t have before.

They’re backed by real knowledge. Each skill ships with curated reference files — methodology, not prompt tricks. Claude loads them on demand so they don’t eat your context window.

They read your actual code. Not your description of it. Findings reference specific files and line numbers.

---

How skills work

your-project/
└── .claude/skills/
    └── /
        ├── SKILL.md          # persona + process + principles
        └── references/       # domain knowledge, loaded on demand

SKILL.md defines who Claude becomes when you invoke the skill. Reference files are the knowledge base. Each skill is self-contained — no shared dependencies.

---

Who this is for

Anyone building things with Claude! Solo devs, small teams, PMs shipping with Claude Code who need expertise they don’t have and are done getting generic AI advice.

---

Contributing

Got a skill backed by real methodology? Not a prompt wrapper — something with actual domain knowledge, tested on a real project.

PR it. Structure: skills/your-skill/SKILL.md + skills/your-skill/references/. Self-contained. No shared deps.

---

License

MIT