| name | skill-name |
| description | What this skill does, leading word first. Then one trigger per branch -- each genuinely distinct situation that should fire it, listed once, no synonym piles. If a trigger collides with a sibling skill, route it: "X is other-skill's job." |
Skill Name
What this skill does and why it exists, in one paragraph. If there
is a one-line test for whether the output is good, state it here.
Gate
Open by sizing the task. When should the agent say "no" or "not
this heavy"? A table mapping task weight to protocol depth beats
running the full ceremony on a typo fix.
Process
- Phase one. End on a completion criterion the agent can check:
done vs not-done must be decidable, exhaustive where it matters
("every X accounted for", not "handle X").
- Phase two. Explain why behind any rule the model might
rationalize away -- reasons adapt, bare MUSTs rot.
- Check. A rubric of concrete, smashable defects. A check the
agent can't fail is decoration.
Anti-patterns
Named failure modes for diagnosis (reference, not steering).
In the process itself, steer positive: state the target behavior
instead of prohibiting the failure.
References
| Reference | When to load |
|---|
| references/example.md | Only some runs need it -- say which |
Inline what every run needs; disclose what only some branches
reach. The pointer's wording decides whether it gets loaded.
