Skill Title
Two or three sentences: what a good result looks like, and what goes wrong when this task is done badly. This framing is what lets the model make sensible judgment calls the steps below do not anticipate.
Workflow
- Gather inputs. What to ask for or read. Which formats are expected.
- Run the deterministic part.
python scripts/<tool>.py <input> --format jsonand why it is better than eyeballing. - Apply judgment. The analytical step. Point to
references/<topic>.mdfor the deep material. - Produce the deliverable. Point to the template in the Output section.
- Recommend next actions. Who to notify, what to hunt, which sibling skill takes over.
Output
Exact template when the deliverable has a fixed shape:
# <Title>
## Summary
## Findings
## Recommended actions
Things that go wrong
- Mistake an experienced analyst warns a junior about, and how to avoid it.
- Another one.
Customization
Edit references/environment.md with your tooling, thresholds, naming conventions, and
escalation paths. Say which fields change the skill's behaviour.