---
name: <short-kebab-name>
description: >
  One or two sentences on what this does, written so a dispatcher can match it
  from the description alone. Include explicit "Use when" triggers.
use_when:
  - <an observable situation, not a vibe: "the task is to publish a doc to the wiki">
  - <another trigger phrasing a person might actually type>
do_not_use_when:
  - <the near-miss case this gets confused with, and what to use instead>
---

# <Skill name>

## Required inputs

List what must exist before this runs. If an input is missing, the skill stops
and asks rather than guessing.

| Input | Where it comes from | If missing |
| --- | --- | --- |
| <name> | <source> | <stop and ask / use this default> |

## Steps

Numbered and deterministic. Each step should be something a reader could do by
hand and get the same result. Avoid "consider" and "as appropriate": if there is
a decision, name the rule that decides it.

1.
2.
3.

## Pass or fail checks

The part most skills skip and the part that makes them trustworthy. Each check
is mechanical: it either passes or it does not, with no judgment call.

- [ ] <check that the output has the required shape>
- [ ] <check that a claim traces to a source>
- [ ] <check the failure mode this skill exists to prevent>

## Notes

Anything a future reader needs: known edge cases, what this deliberately does
not handle, and the reason for any rule that looks arbitrary.