Skill Activation

Skills auto-activate when Claude recognises a matching use case. Each skill ships with a description in its frontmatter — that text is what Claude reads when deciding whether the skill is relevant to your current request.

How It Works

When you ask Claude to do something, it scans the descriptions of every loaded skill and matches your request to the closest one. If a skill's description names the situation you're in, Claude invokes it without you typing a slash command.

You: "I need to plan the architecture for a multi-file change"
  ↓
Claude scans descriptions across loaded plugins
  ↓
Finds specs:plan, whose description names this scenario
  ↓
The skill activates and Claude follows its protocol

When a Skill Doesn't Activate

If you expected a skill to fire and it didn't, two things to check:

1. Was the plugin installed and loaded? Run /plugin and confirm the skill's parent plugin is listed.
2. Was your request specific enough? Skills match observable triggers — "fix this" rarely activates anything; "review this PR for security issues" tends to.

You can also invoke a skill directly. Most skills surface as /<plugin>:<skill>, e.g. /dev:review. Direct invocation skips the description match entirely.

Related

• Component Model — Commands, skills, and agents
• Specs Workflow — Example: specs skills auto-activate by phase
• Plugin Design — For plugin authors writing new skills