Claude Picked Up a Skill Called "log-leave" — Can It Approve Leave Requests?

The video starts from a guess: that once Claude picks up a skill called log-leave, it can decide whether to approve an employee's leave request. It says that is wrong. Claude follows a written checklist, and the video walks through what that file is and what it can and cannot do.

Nothing in the model changes

A skill named log-leave sounds as if Claude itself can now decide whether to approve someone's leave. But nothing in the model changes. Delete the skill's folder, and Claude does not lose any approval authority, because there was none to begin with. It simply stops following that one checklist for logging a leave request.

What a skill actually is

A skill is a folder holding a single file, SKILL.md. For log-leave, that file contains instructions for logging a leave request step by step, written in plain language that Claude reads before it starts. Claude reads it top to bottom and works through it in the order written, with no branching unless the file itself says to branch.

That makes a skill a specification, not new judgment.

The payoff and the limit

The payoff is that the same kind of log entry is produced for every leave request, every time. The limit is anything outside the written steps. Whether the leave even qualifies is not something the checklist covers, and Claude has no special opinion about it. Log-leave never gave Claude authority to approve a request; it only guarantees that each time it runs, Claude reads the same file and logs the leave the same way.

What watching it work does not prove

The video adds a caution about how to read Claude's output. Watching Claude log a leave request cleanly does not prove it judged the situation correctly. A checklist can be followed to the letter and still miss what it was never asked to check. And watching Claude flag an entry as incomplete does not prove something is wrong; it may be a case the checklist does not cover. Anything outside those steps is still on you.

What the video does not claim

The video does not describe what an actual log-leave file instructs, because that file lives in a collection not reachable from the build. The name is used only as an example of a skill-shaped folder. What it shows is what is generally true about how any Claude skill works.

Key takeaways

• A skill named log-leave does not give Claude the authority to approve leave.
• Deleting the skill folder removes the checklist, not any approval power.
• A skill is a folder with one SKILL.md file of plain-language steps.
• Claude reads the file before starting and follows the steps in order.
• A skill is a specification, not new judgment.
• The benefit is the same log entry every time; the limit is anything the file never wrote down.
• A cleanly logged entry does not prove the request was judged correctly, and a flagged one does not prove a problem.

Try it yourself

Pick one record you log the same way every time, such as a time-off request, an expense entry or a status update. Ask Claude to write a SKILL.md for it in plain language with ordered steps. Then have Claude read it back and walk through exactly what it will log, before it logs anything. Try it today on a record you actually fill out over and over.