Resources / Skills

AI Skills Library.

122 curated Claude Code skills covering frontend, SEO, architecture, testing, devops, 3D, security, and AI agents. Each skill includes capabilities, use cases, and project-specific notes. Plus slash command reference, settings.json schema, and git hooks.

122

Total skills

Curated Skill Library

122 skills available

Format Reference

Skill File Format

Skills live at .claude/skills/<name>.md — YAML frontmatter declares the name and description; the Markdown body becomes the skill's system prompt when invoked.

---
name: deploy
description: Deploy the current branch to production via Railway
---
# Deploy Skill

Run `railway up --environment production` for the current project.

## Steps
1. Check for uncommitted changes (git status)
2. Confirm the target environment with the user
3. Run: railway up --environment production
4. Tail deployment logs until success or failure
5. Verify health endpoint responds 200

## Requirements
- RAILWAY_TOKEN must be set in environment
- railway CLI installed globally

Git Hooks Integration

Wire Claude Code into Git Hooks

Run Claude Code skills automatically at key git lifecycle events. Install hooks in .git/hooks/ or use Husky for team-wide hooks via .husky/.

pre-commitReview staged files

Runs /review on staged TypeScript files. Blocks commit if critical issues are found.

#!/bin/bash
# .husky/pre-commit
set -e
STAGED=$(git diff --cached --name-only --diff-filter=ACMR | grep -E '\.(ts|tsx)$' || true)
if [ -z "$STAGED" ]; then exit 0; fi

echo "Running Claude Code review on staged files..."
echo "$STAGED" | xargs -I{} claude --print "/review {}" 2>&1 | tee /tmp/claude-review.txt

if grep -q "severity: critical" /tmp/claude-review.txt; then
  echo "Claude Code found critical issues. Fix before committing."
  exit 1
fi

commit-msgValidate commit message

Validates that commit messages follow Conventional Commits spec via Claude.

#!/bin/bash
COMMIT_MSG=$(cat "$1")
if echo "$COMMIT_MSG" | grep -q "^Merge"; then exit 0; fi
RESULT=$(claude --print "Validate this commit message follows Conventional Commits. Reply PASS or FAIL.\n\nMessage: $COMMIT_MSG" 2>/dev/null)
if echo "$RESULT" | grep -q "^FAIL"; then echo "$RESULT"; exit 1; fi

pre-pushSecurity audit before push

Runs /audit on the full codebase before pushing. Blocks if HIGH or CRITICAL issues found.

#!/bin/bash
set -e
BRANCH=$(git rev-parse --abbrev-ref HEAD)
echo "Running Claude Code security audit before push to $1/$BRANCH..."
claude --print "/audit" 2>&1 | tee /tmp/claude-audit.txt
if grep -qE "(HIGH|CRITICAL)" /tmp/claude-audit.txt; then
  echo "Security audit found issues. Resolve before pushing."
  exit 1
fi

Configuration Reference

~/.claude/settings.json Schema

Controls Claude Code permissions, hooks, environment variables, and model selection. Place at ~/.claude/settings.json for global config or .claude/settings.json for project-level overrides.

permissions.allow

Glob patterns for commands and file operations Claude can execute automatically. Use Bash(*), Read(**), Write(src/**) patterns.

permissions.deny

Patterns ALWAYS blocked regardless of context. Use for irreversible operations you never want automated.

hooks.PreToolUse

Shell commands run BEFORE each tool call. Return non-zero exit code to block the call. Receives CLAUDE_TOOL_INPUT env var.

hooks.PostToolUse

Shell commands run AFTER each tool call completes. Great for running typecheck after every Write.

hooks.Stop

Shell commands run when Claude finishes responding. Use for notifications: alerts, Slack, terminal bells.

env

Key-value pairs injected as environment variables into every Claude Code session. Useful for NODE_ENV, log levels, feature flags.

model

Default model for primary responses. Example: "claude-sonnet-4-6".

smallFastModel

Faster/cheaper model for classification, routing, and quick tasks within the same session.

{
  // Permissions — what Claude Code can and cannot do automatically
  "permissions": {
    "allow": [
      "Bash(git log:*)",
      "Bash(npm run *)",
      "Bash(npx tsc --noEmit)",
      "Read(**/*)",
      "Write(src/**/*)"
    ],
    "deny": [
      "Bash(rm -rf /)",
      "Bash(git push --force *)"
    ]
  },

  // Hooks — shell scripts run at lifecycle events
  "hooks": {
    "PreToolUse": [{
      "matcher": "Bash",
      "hooks": [{ "type": "command", "command": "echo 'Running: $CLAUDE_TOOL_INPUT'" }]
    }],
    "PostToolUse": [{
      "matcher": "Write",
      "hooks": [{ "type": "command", "command": "npx tsc --noEmit 2>&1 | head -20" }]
    }],
    "Stop": [{ "hooks": [{ "type": "command", "command": "notify-send 'Claude Code' 'Done'" }] }]
  },

  // Inject env vars into every Claude Code session
  "env": { "NODE_ENV": "development", "NEXT_PUBLIC_LOG_LEVEL": "debug" },

  "model": "claude-sonnet-4-6",
  "smallFastModel": "claude-haiku-4-5-20251001"
}

Best Practices · Template

CLAUDE.md Template

Place CLAUDE.md at your project root. Claude Code reads it on every invocation. Define stack, rules, common tasks, and architecture decisions here.

# Project: My App

## Stack
- Next.js 16, React 19, TailwindCSS v4, TypeScript strict
- Tests: Vitest + Testing Library

## Critical Rules
- Static export only (no API routes, no SSR)
- Commit only when explicitly asked
- No console.error — use reportError()
- All strings via useTranslations()

## Common Tasks
- Dev: npm run dev
- Build: npm run build
- Typecheck: npm run typecheck
- Tests: npm run test