One install gives your agent the second half of the prompts-to-skill pair: it takes a skill that is already drafted and tuned, runs its trigger test cases and records what passed, sets the version and CHANGELOG, builds the zip, writes the README, and ends with a per-platform install checklist. A failing test sends the skill back to tune-skill instead of getting a quiet edit here.
install
Detects Claude Code, Cursor, Codex, Qoder and more, then writes the skill into the right folder. Install tune-skill alongside it: precondition 1 reads that validator, and this repo does not carry it.
npx skills add arm-pz/prompts-to-skill-release --skill prompts-to-skill-release
Quality record
Version, date and limits are read from the repo on every sync, and the check runs every night. Real-world runs are listed only with their evidence.
the run
SKILL.md holds the eight steps, references/release-steps.md the detail, readme-template.md the skeleton step 6 fills in. It packages and documents one thing and stops there: drafting belongs to prompts-to-skill, auditing a skill that exists to tune-skill, naming and the GitHub listing to findable.
It stops and reports unless all three hold. A gate marked "written, not run" counts as a warning, not a pass.
Every case in evals.json and the tune-skill table gets one of four results: pass, fail, reviewed-only, not-run.
The first release is 0.1.0. An existing version field is left alone unless you ask for a bump.
The skill folder is the archive root, so unzipping gives skill-name/SKILL.md.
Three prompts that should trigger it, in your own words, and one that should not although it shares keywords.
The handover. Step 8 runs only if you ask for it, and then hands the live repo to findable for the name decision and listing.
house rules
The report carries counts for pass, fail and reviewed-only, and a not-run case is named in its first line. The rule is one the skill sets on itself: a failed test means the skill is not ready, and the report says so rather than tidying it away.
Step 2 sorts every failure into a description problem or a body problem and sends both back to tune-skill. Nothing about the skill's behaviour changes during a release. The skill that authors the draft is prompts-to-skill; this one only ships it.
Publishing is step 8 and runs only on request. The archive is verified with a listing before anyone calls it packaged, and the discovery check is recorded as a read of the description, not as proof the skill surfaced in a live session.
Free, open, and it never edits the skill you wrote. You decide what gets published.
Install prompts-to-skill-release Read the source