Claude Skills Fire by Description, Not by Name

A common assumption is that you have to type a skill's exact name to switch it on. The video says you do not. You describe the task, and Claude checks your wording against the description written in the skill's file. When your words land inside that description, the skill fires. When they do not, the skill stays off, and Claude still answers, only without that file's playbook.

A skill is a folder with one file

A Claude skill is a folder holding a single file, SKILL.md. Near the top of that file sits a plain paragraph that says exactly when the skill should be used. That paragraph is what does the matching. Nobody has to name the skill to activate it. The video's example is a request like "help me write a rule for Hookify": it never names any skill, and the matching skill still fires.

Claude reads your request against the paragraph

According to the video, Claude reads your request against the skill's description word for word. There is no menu, no button, and no name required. The worked example is a skill called writing-hookify-rules. Its real description says it fires for four phrases: "create a hookify rule," "write a hook rule," "configure hookify," or "add a hookify rule."

The video asks the viewer to watch this anchor. Type a request that lands inside that wording and the match fires the skill. Claude then opens the file and follows it step by step.

Same wording in, same skill out

Ask again, this time for something like adding a rule to Hookify that blocks a dangerous command, and the same match fires again. The video's phrasing is that the same wording in produces the same skill out, every time.

When nothing matches

The contrast case is a request unrelated to the description. The video uses "explain YAML in general." None of that wording appears in the skill's paragraph, so the skill stays off. Claude does not refuse or get stuck; it answers from what it already knows, just without this file's playbook. A request that does not match simply does not fire that particular file.

What happens after a match

Once the skill fires, Claude follows exactly what the file says and nothing more. The video's carry-out is that a skill fires when your words match its description, not when you name it. The video also keeps its claim narrow. It states the mechanism shown by the skill's own description field and does not say how the matching is computed internally.

Key takeaways

• You do not need to type a skill's name to use it; you describe the task.
• A skill is a folder with one file, SKILL.md, and a plain paragraph near the top says when to use it.
• Claude compares your request against that paragraph, with no menu, button, or name involved.
• The writing-hookify-rules skill fires for phrases such as "create a hookify rule," "write a hook rule," "configure hookify," and "add a hookify rule."
• Similar wording produces the same skill every time.
• An unrelated request, such as explaining YAML in general, leaves the skill off while Claude still answers.
• After a match, Claude follows what the file says and nothing beyond it.

Try it yourself

Ask Claude: "I want to add a hookify rule that blocks rm -rf commands. Read the writing-hookify-rules skill first, and walk me through what you'll do before you do it." Notice that you never name the skill in the task itself. Asking Claude to explain its plan before acting shows whether the match actually landed on the right skill. If you have your own skill, rewrite its description using phrases people would really say and try the same test.