Skill ports
Skills trending on GitHub, X and builder communities are ranked by how many installs they would likely get, checked against five gates, packaged as SKILL.md, tested on representative prompts, and launched on a steady cadence.
Selected skill
Ranked candidates
How to read this: rank comes from predicted 30-day installs. Gate pills, in order: License · Creator permission · Attribution · Transport · Sandbox. Green passes, amber is something we can fix (ask the creator, legal review), red stops the port. Click a row for its port plan.
How candidates are ranked
What predicts installs
How to read this: a regression of log installs on log signals. Coefficients are elasticities: +10% stars in the last week ≈ +% installs, all else equal.
Momentum beats total stars
How to read this: rank correlation between predicted and actual installs for ported skills the model never saw, compared with ranking by total GitHub stars. Spotting skills before they peak means weighting this week's momentum, not lifetime popularity.
Launch cadence
How to read this: grey bars are past launches; blue bars are the plan. The plan launches what has passed QA first, then what is ready to port (one week to package and test), at a week.
How a port ships
1. Package
One folder per skill: SKILL.md (instructions + attribution), listing.json (marketplace copy), qa_log.md (every test prompt and result).
2. Pull request
The folder goes to the marketplace repo as a PR. The GitHub export workflow runs the license, attribution and listing-template checks; a failing check blocks the merge.
3. Launch
Merge exports the skill to the marketplace. Launch notes credit the creator by handle and link the source repo, the same day.
Future scope: first-party plugins
First-party plugins use the same path. A subject-matter expert writes the SKILL.md recipe, we run the same QA prompts, and the listing goes through the same template checks. Only the attribution line changes ("Built by the Marketplace team with team name").