⭐ If you find these skills useful, star this repo — it helps others discover them!

Practical Claude Code skills born from real production projects, not templates.

The fastest way to understand an unfamiliar codebase with Claude Code.

npx skills add Wubabalala/claude-skills@project-onboarding

Then in Claude Code:

/onboard

You join a new project. You spend hours reading scattered docs, guessing build commands, figuring out which configs matter. Existing "onboarding" tools dump a directory listing and call it done.

Phase 0: Detect & Scan
  ├── Finds existing AI files (CLAUDE.md, .cursorrules, AGENTS.md, .windsurfrules...)
  ├── Scans project structure coarse-to-fine (type → scale → key info)
  ├── Cross-checks docs against code — surfaces stale commands, wrong ports, outdated claims
  └── You decide: keep / patch / rebuild / skip each file

Phase 1: Generate Core Docs
  ├── CLAUDE.md — every claim backed by a source file path
  ├── Per-module CLAUDE.md (monorepo auto-detected)
  ├── docs/OVERVIEW.md — one-page project navigator
  └── Security scan blocks writes if credentials detected

Phase 2: Deep Dive (optional)
  ├── Recommends dimensions based on what it found (deploy scripts → ask about deployment)
  ├── Asks "WHY was it designed this way?" — captures what code can't tell you
  └── Writes memory files with clear fact vs user-knowledge separation

Code tells you what exists. It doesn't tell you:

• Why there are two auth systems (legal required SSO for one, JWT for the other)
• Why deploys must run macOS before Windows
• Which config change broke prod last month
• That the test script in package.json hasn't worked since the migration

This skill extracts facts from code, then asks you the questions that matter.

After running /onboard on a microservices project:

Scan Results

Project type: Java (Spring Cloud) + Next.js + Python (FastAPI)
Scale: 1,247 files, 7 modules
Build tools: Maven, pnpm, uv

Existing AI files:
  CLAUDE.md (root) — Covers: build commands | Missing: service topology, deployment
  .cursorrules — Covers: code style | Conflict: claims React, project uses Vue

Conflicts: 1 found
  .cursorrules says "React", package.json has "vue": "^3.4" → Which is correct? [Code] / [Doc]

Recommended actions:
  1. CLAUDE.md — [Keep & enhance] / [Patch conflicts] / [Rebuild] / [Skip]?
  2. .cursorrules — [Merge into CLAUDE.md] / [Ignore]?
  3. docs/OVERVIEW.md — [Generate] / [Skip]?

• Code is fact, documentation is claim — conflicts surfaced, user decides
• Source-safe — never modifies source code or configs; only writes doc files
• User confirms all writes — nothing touches disk without approval
• Template is maximum, not minimum — sections skipped if not applicable
• Ask "why", not "what" — code scanning is commodity; asking the right questions is the value

Pragmatic code review & pre-push quality gate, with a machine-readable verdict.

npx skills add Wubabalala/claude-skills@code-review

Then in Claude Code:

/code-review

Risk-based code review that catches real bugs (not style nitpicks) and produces output that both humans and machines can trust:

Universal Layer 1 checks (always)
  ├── P0 Security (hardcoded secrets, SQL injection, XSS, auth removal)
  ├── P0 Transaction Safety (cross-table atomicity, rollback-only catches)
  ├── P0 Logging Secrets Leakage (runtime print + client-side persistence)
  ├── P1 Logic Bugs (crashes, race, return/build mismatch)
  ├── P1 API Input Validation (NPE on missing fields, parse-without-try)
  ├── P2 Robustness (N+1, empty catch, ORM pitfalls)
  └── P3 Maintainability

Project Layer 2 checks (optional, per-project)
  └── .claude/review-checklist.md — auto-generated from your CLAUDE.md and docs

Output (every report ends with a machine-readable sentinel):
  <!--CODE_REVIEW_GATE_BEGIN-->
  REVIEW_GATE=PASS|FAIL
  REVIEW_P0_COUNT=<int>
  ... (P1/P2/P3 counts)
  REVIEW_VERSION=2.1
  <!--CODE_REVIEW_GATE_END-->

Unknown --mode= values fall back to interactive.

Without a stable machine-parseable verdict, any pre-push hook that greps [ NEEDS FIXES ] from natural-language output will eventually misparse — false positives block clean code, false negatives let regressions through. The sentinel block fixes that:

• HTML-comment wrapped (invisible in rendered markdown, greppable by hooks)
• Fixed schema, fixed field names, fixed position (always last)
• v2.2 hook (planned) will consume ^REVIEW_GATE=(PASS|FAIL)$ directly

• v2.1 (this release): machine-readable verdict layer + secrets-in-logs P0
• v2.2 (planned): pre-push git hook (sentinel-driven, read-only, install/uninstall scripts)
• v2.3 (planned): architecture-traps two-way binding + 3-dimensional rule metadata

Browser automation that goes beyond recording — render diagrams, automate interactions, export screenshots.

npx skills add Wubabalala/claude-skills@playwright-web-automation

• Web interaction — fill forms, click buttons, navigate multi-step flows, handle login state
• Diagram rendering — Mermaid, custom HTML/SVG → high-res PNG export
• Canvas export — extract rendered content from Canvas elements (even inside iframes)
• Screenshot automation — full page, element-level, or high-DPI (2x) capture

SKILL.md (decide which path → skeleton steps)
  ├── references/render-to-image.md    (setContent + CDN rendering patterns)
  ├── references/export-strategies.md  (screenshot / Canvas / download)
  ├── references/iframe-and-canvas.md  (cross-origin iframe, contentFrame)
  ├── references/wait-strategies.md    (condition-based waiting, avoid sleep)
  ├── references/troubleshooting.md    (common errors + fixes)
  ├── templates/skeleton.mjs           (starter script with both paths)
  └── examples/
      ├── diagramgpt.mjs              (full: iframe + follow-up forms + Canvas)
      └── mermaid-render.mjs           (full: CDN Mermaid → batch PNG)

• Two paths, one skeleton — recording is optional, not mandatory
• Headless by default — headless: true for batch/render tasks, false for debugging
• Reference on demand — main doc stays under 120 lines; details loaded when needed
• Real examples — both examples are production-tested, not toy code

More skills coming. Each one is built from real production workflows.

⭐ Star this repo if it saved you time · 💬 Issues & ideas welcome

MIT