Agent Skill Launch

The marketer's playbook for shipping an AI-agent skill (SKILL.md) that gets loaded and installed — not just committed. Built from a teardown of skills that took off in August 2026 and the ones that didn't.

When to Use

• About to ship a SKILL.md and want it to reach agents.
• Want a skill/product to earn installs, not just sit in a repo.
• Asking "how do I get on skills.sh" or "why did that skill go viral".

The three levers

1. Distribution is install-driven, not crawled. A repo appears on the skills.sh leaderboard only after someone runs npx skills add owner/repo — that install fires telemetry, and installs (not stars) decide ranking. Seed your own skill once so it becomes discoverable.
2. The description fires the skill. Claude Code, Cursor, and skills.sh all read the frontmatter description to decide when to load the skill. Write it LONG, naming trigger situations ("Use when the user asks for / wants to …"), never a capability summary.
3. One memorable hook. Skills that get cited and reposted have a single quotable rule — not a feature list. Example that worked: "prose is authored, counts are measured" plus a drift counter that fails CI when the map rots. Pick one line someone can quote.

How to write the SKILL.md

• Frontmatter: name (lowercase-hyphen) + a description written as the firing trigger (above).
• Body: intro (what it does AND does not do) → When to Use (+ don't-use counter-triggers) → Prerequisites → How to Run → Procedure (numbered, each step with a completion criterion) → Pitfalls → Verification.
• Keep SKILL.md scannable. Push detail into references/, executables into scripts/, files-to-copy into assets/.