ScottDuncanAI Skill Name
One or two sentences on what a good result looks like and who judges it. Naming the reviewer — "a process engineer who will spot a backwards arrowhead" — sets the bar far better than an adjective like "high quality".
Non-negotiable output rules
The handful of things that make the output wrong if violated. Keep this short; if everything is non-negotiable, nothing is.
- …
- …
Workflow
1. Read the request and fill the gaps
What is almost always underspecified? List the questions worth asking, and cap them (e.g. "ask at most 2–3, and only for things that change the answer"). For everything else, make a defensible assumption and state it afterward.
2. Plan before producing
The step that prevents rework.
3. Produce
Point at scripts/ and references/ here rather than restating their contents:
scripts/<module>.pyprovides the primitives. Read it before writing anything and use it rather than reinventing. Readreferences/<topic>.mdfor the conventions.
4. Check the output and fix it
The step that separates a passable result from a professional one. Render it, run it, or validate it — then actually look at the result and iterate. Say how many rounds to expect and list the defects that are only visible on inspection.
5. Deliver
Where the output goes, and what to state alongside it: assumptions made, anything that couldn't be done and why, and any concern noticed along the way. Flagging concerns is usually a large part of the value.
What this skill does not do
State the limits plainly, and say so to the user if their framing suggests they expect more. Everything in this repo is conceptual, engineering-support output — not issued-for-design work, not hazard analysis, not equipment sizing.