Automatic and Explicit Invocation

Claude chooses, or you ask by name

There are two ways a skill gets used. Automatically: Claude reads your request, compares it with the available skill descriptions and, when one fits, loads that skill and follows it. This is why descriptions matter so much. Explicitly: you can ask for a skill by name in your message, and in current Claude Code versions you can also invoke skills as slash commands (/skill-name), optionally with arguments. Explicit invocation is useful for skills that perform actions (deploy, release) which you want to trigger only on purpose, and some tools let a skill be marked so the model does not invoke it by itself; check the documentation for the options in your version. If a skill does not trigger, check that the folder and file names are right, the frontmatter parses, and the description matches how you phrased the request; if it triggers too often, make the description narrower.

Troubleshooting a skill that does not trigger

Work down the list.

1. path correct?       <scope>/skills/<name>/SKILL.md   (file is exactly SKILL.md)
2. frontmatter valid?   starts with --- on line 1; name + description present; no tabs/odd quotes
3. name == folder?      lowercase-hyphen name that matches the folder
4. description match?   does it contain the words people actually use? does it say WHEN to use it?
5. conflicts?           another skill with a similar description or the same name?
6. restart/reload?      start a new session if your version only scans at startup
7. try explicit call    ask for it by name or /name to separate "not found" from "not chosen"

Name the skill in a hard case

If automatic selection is flaky, ask for the skill by name while you tune the description.

Quick check: A skill never triggers automatically. What is the most useful first check?

• Whether the model is large enough
• Whether its description contains the words people actually use and says when to use it
• Whether the repo has a licence
• The colour of the terminal