Menu
Drafts or revises a Skill Bundle's output/SKILL.md from its design.md. Use when handed a bundle's design.md (and optionally prior review revise notes) and asked to produce or update output/SKILL.md.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Cut a skillmaker-studio release — version bump PR, v-tag push, release.yml build + npm publish, verification. Use when a maintainer asks to release a new version of skillmaker-studio to GitHub Releases and npm.
Drafts or revises a Skill Bundle's output/SKILL.md from its design.md. Use when handed a bundle's design.md (and optionally prior review revise notes) and asked to produce or update output/SKILL.md.
Researches a skill's topic and writes research/notes.md before design.md or SKILL.md is drafted. Use when handed a bundle's topic (bundle.json name/oneLiner and any existing design.md content) and asked to research it and write research/notes.md.
Decompose a decided plan, spec, issue, or design conversation into dependency-ordered tracer-bullet tickets written as local markdown files. Use when asked to "break this into tickets," "create implementation tickets," or "decompose into work items" — any time a decided scope needs slicing into executable vertical-slice tickets with blocking edges.
Design, evaluate, and ship agent skills with receipts using Skillmaker Studio. Use when the user asks to set up Skillmaker in this repo, create/adopt/run/grade/ship a Skill Bundle, open the board, or otherwise work with the skillmaker CLI (init, new, start, run, grade, ship, publish, adopt).
Add the project's canonical SPDX license header to source files that are missing one. Use when asked to add a license header to a file or directory, e.g. "add the license header to src/foo.ts" or "make sure every file in src/ has our license header."
| name | william-draft-skill-md |
| description | Drafts or revises a Skill Bundle's output/SKILL.md from its design.md. Use when handed a bundle's design.md (and optionally prior review revise notes) and asked to produce or update output/SKILL.md. |
You are drafting output/SKILL.md for a Skill Bundle, working in a sandbox
seeded with the bundle's current design.md (and output/SKILL.md, if one
already exists). Follow these steps in order.
Read design.md in the current directory.
If it does not exist, or its ## Intent and ## The workflow sections
are empty or still just the scaffold's HTML comments with no real
content: stop, write nothing. Do not create output/SKILL.md. End
your final message with a plain statement that design.md doesn't have
enough content yet to draft a SKILL.md. A fabricated skill is worse than
no skill.
Check your prompt for a "REVISE NOTES:" section.
If present, a human reviewer already looked at a previous
output/SKILL.md draft and is asking for something specific. Treat the
revise notes as your primary instruction for this pass, on top of (not
instead of) staying faithful to design.md.
Check whether output/SKILL.md already exists.
If it does, treat it as a first draft to revise, not something to
discard. Preserve any part that still matches design.md's current
## Intent / ## The workflow. Rewrite only what has drifted, or what
the revise notes call out.
Write (or update) output/SKILL.md:
name (the bundle's slug, kebab-case, from
design.md's frontmatter bundle: field) and description (one to
two sentences, derived from design.md's ## When to use / triggers
section -- this is what an agent's tool-selection layer reads to
decide whether the skill is relevant, so make it concrete and
trigger-shaped, not vague).design.md's ## The workflow into a concrete numbered
procedure. Explicitly carry over every "must never" / "always stop
and ask" constraint listed in design.md's ## Failure hypotheses
table -- each one exists because someone already found a way this
skill goes wrong; dropping one silently reintroduces that failure.Stay in scope. Only create or edit design.md and
output/SKILL.md. If you notice something that belongs in research/
or evals/, mention it in your final message -- do not edit those
directories yourself.
Stop once output/SKILL.md reflects design.md. Do not run the
drafted skill, do not write eval fixtures, do not change the bundle's
stage. Those are separate, human-gated steps.