For each host you ship to: will this SKILL.md actually load — LOADS, LOADS-EMPTY-METADATA, or silently DROPPED — and exactly why not?
One in eight skill repos ships a SKILL.md a host cannot parse (330 of 2,621 repos, 12.6%, in a measured census of 336,285 files), hosts drop it silently, and nothing tells the author. A strict linter cannot tell you what a host will do — hosts differ in leniency, limits, and unknown-field handling, and one silent-drop cause (the shared listing budget) exists only at the level of your whole skill set. This service grades a skill against each covered host's loader semantics and returns, per host, LOADS / LOADS-EMPTY-METADATA / DROPPED with the exact cause named (parse error with line, name≠directory, description bytes vs that host's limit, BOM, nesting, unknown-field handling). Marketplaces and CI call the publish gate before listing — a DROPPED verdict files a fix-ready notice to the author. Operators run the installed-fleet census to find the skills they believe are in force that are in fact dropped or budget-evicted in the running configuration — including the safety or procedure skill that never loaded.
Cross-host loadability verdict $0.01 per skill · Publish gate + author notice $0.99 per listing · Installed-fleet census $4.99 per sweep · Fleet pass $19.99/mo (4 census sweeps + 2,000 cross-host verdicts per month; publish gates metered at $0.99; beyond the allowance metered per call). One free single-skill cross-host check per operator, ever.
Stated plainly: host personas are labeled emulations of documented loader behaviour, not verified real-host binaries; willingness to pay is unmeasured; the fleet pass is a stated allowance, never unlimited (see pricing).
Support: support@textforagents.com