Claude Code Skills

Appearance

Frequently Asked Questions ​

Basics ​

Q: How is a Skill different from a normal Claude Code conversation? ​

Skills load domain-specific expertise and workflows. For example, /review follows a structured "correctness→security→performance→maintainability" process, while a normal conversation might only focus on the aspect you mention.

Q: Can I use multiple Skills at once? ​

Not recommended. Skills change Claude's behavior mode; simultaneous use causes interference. Use one at a time, switch when done.

Q: Do Skills remember previous conversations? ​

No, Skills don't persist across sessions. Use /handoff to create a handoff document for cross-session continuity.

Q: Where do I put custom Skills? ​

  • Project-specific: .claude/skills/
  • Global: ~/.claude/skills/

Triggering ​

Q: I said the trigger words but the Skill didn't activate? ​

Possible reasons:

  1. Description too vague, overlapping with other Skills
  2. Another Skill is already active
  3. Use /skill-name slash command — most reliable

Q: A Skill auto-triggered but I don't want it. How to exit? ​

Say "exit current Skill" or trigger a different Skill.

Q: How to prevent false triggers? ​

Limit trigger conditions in settings.json, or use only /command based triggering.

Efficiency ​

Q: Do Skills consume more tokens? ​

Skills load prompt instructions on activation (one-time cost). The efficiency gain far outweighs this cost. Use /caveman to further reduce tokens.

Q: Which Skills are used most often? ​

By frequency:

  1. review — After every change
  2. diagnose — When something breaks
  3. tdd — Writing new features
  4. simplify — After review

Q: How to manage context in long sessions? ​

  1. Use caveman to reduce dialog tokens
  2. Use handoff between phases, start fresh session
  3. Use loop for background periodic checks

Custom Development ​

Q: How do I debug my custom Skill? ​

  1. Test basic trigger activation
  2. Test with various phrasings for false triggers
  3. Verify output format
  4. Observe edge-case behavior

Q: Can Skills call external APIs? ​

Yes. Skills can use WebFetch for external APIs. Configure MCP tools for database access.

Q: How long should a Skill prompt be? ​

  • Simple: 20-50 lines
  • Medium: 50-150 lines
  • Complex: 150-300 lines
  • Over 300 lines? Move details to resources/

Troubleshooting ​

Q: /diagnose didn't find the issue. What now? ​

Provide more detail: exact error messages, reproduction steps, environment (OS, versions). Try narrowing manually and re-running diagnosis.

Q: I can't write the TDD test. What should I do? ​

The design might need adjustment. Easily testable code is usually better designed. Use /prototype to validate the approach first.

Q: The generated PRD is inaccurate? ​

PRDs extract from the conversation. If the conversation is incomplete, the PRD will have gaps. Always review and supplement with business context.