Frequently Asked Questions

If your tasks are one-off and general, base Claude is perfect. But Skills shine when you: - Repeat similar tasks often (saves explaining context every time) - Need to follow specific company workflows - Require deep domain expertise not in Claude's training - Want consistent results every time Think of it this way: base Claude is a smart generalist, Skills add specialist knowledge.

Honestly? No. We've collected 200,000+ skills from the community - quality varies widely. We haven't personally tested each one. That's why we're building: - Rating and review system - Community feedback mechanisms - Quality filters and badges For now, treat Skills like open-source code - review before use, check GitHub stars, read reviews.

Look for these signals: ✓ High GitHub stars (community validation) ✓ Active maintenance (recent commits) ✓ Clear documentation in SKILL.md ✓ Real-world use cases in README ✓ Comes from domain experts, not just AI-generated Coming soon: User ratings and reviews with real scenarios.

Yes, treat Skills like any open-source code: ⚠️ Skills can contain scripts (Python, Bash, etc.) ⚠️ Always review the code before installation ⚠️ Check the author's reputation on GitHub ⚠️ Prefer skills from trusted sources We don't scan for malware (yet). This is community-driven content - use common sense and due diligence.

Most skills work everywhere: ✅ Regular Skills (Python/JS/Markdown) → Web + Desktop + CLI ⚠️ Skills using MCP tools → Desktop + CLI only (not Web) Check the skill's detail page for platform compatibility tags.

Batch installation is coming soon! Currently: - Use /plugin marketplace for skills with marketplace.json - Or download ZIP and manually copy to ~/.claude/skills/ We're working on: - Batch download by author - Batch download by category - Auto-install related skills

These are all types of Claude Code plugins: • Skills (SKILL.md) - Knowledge and workflows • Agents (AGENT.md) - Autonomous task executors • Commands (COMMAND.md) - Slash commands (/foo) • Hooks - Event triggers (pre-commit, etc.) Many repositories contain multiple types bundled together.

Skills use progressive disclosure: 1. Claude initially sees: skill name + short description (tiny tokens) 2. If potentially relevant: Claude reads full SKILL.md (more tokens) 3. If needed: Claude loads references/ and assets/ (even more tokens) This saves tokens vs. loading everything upfront (like MCP does).

Absolutely! Once installed, skills are just files on your system. You can: ✓ Edit SKILL.md to customize workflows ✓ Add your own best practices ✓ Modify scripts in scripts/ directory ✓ Fork on GitHub and share your version Just respect the original license terms.

We auto-index skills from GitHub! To get indexed: 1. Create a GitHub repo with SKILL.md file 2. Add proper frontmatter (name, description) 3. Add topic: claude-skills or claude-code-skill 4. Wait for our next sync (runs daily) Submission system coming soon for manual additions.