One install runs three skills in order: Draft-Skill confirms intent, checks siblings and writes the SKILL.md; Tune-Skill makes the description fire, resolves overlaps and validates with check_skill.py; Findable-Polish audits the discovery surfaces and polishes the copy. Each phase gates the next — nothing advances on “looks fine”, and no public change runs without your go-ahead.
install
The install line from the repo README. The pipeline needs draft-skill, tune-skill and findable-polish installed; a missing one is named and the run stops.
npx skills add arm-pz/skill-launch --skill skill-launch
the pipeline
Each phase ends in an artifact the next one consumes: a drafted folder, a validated skill with zero FAILs, a polished set of discovery surfaces. The final report states which checks ran and which publication commands are waiting on you.
Intent confirmed before anything is written; siblings checked so the new skill has one clear owner of its request.
Tune-Skill audits and edits, and the folder must pass check_skill.py with zero FAILs to move on.
Findable-Polish derives the queries and polishes README, About and site copy; the launch kit hands over copy-paste commands.
An existing skill that needs work runs through tune-and-polish, not this pipeline. A prompt list rather than an idea is prompts-to-skill territory. Nothing public changes without confirmation — that rule outlives every convenience.
ground rules
A skill written from a misread idea is a repo you have to delete later. Phase one asks what it is for and who should never trigger it, then writes.
The validator decides when phase one is done. Descriptions are measured against the checklist, and a FAIL sends the skill back to tuning instead of forward to polish.
Repos, releases and listing edits are drafted with exact commands and run only after confirmation. The pipeline prepares the launch; you pull the trigger.
Free, open, and every launch traces back to a validated file.
Install skill-launch Read the source