Keeping a catalog of bot definitions valid when the platform moves
A catalog of installable bot definitions is only useful if the definitions still work. Platforms change their schemas without asking.
Every listing is a complete definition: identity, skills, routines and the boundaries the bot works inside. People copy it and run it. If a field is renamed upstream, a listing that looks fine becomes one that fails on paste, and the catalog quietly rots.
So validation runs on a schedule rather than at submission. Every definition is checked against the platform's current schema, and anything that no longer validates is flagged, redrafted against the new shape, and held for review before it goes back on the shelf.
The same check runs over the human-readable half. A description that no longer matches what the definition does is a worse failure than one that will not install, because nobody notices.
What runs now: scheduled schema validation across the catalog, automatic redrafting of definitions that break, a review queue before anything republishes, and a consistency check between what a listing says and what it does.
What it taught us: people read the whole definition before they trust an agent, including the part that says what it is not allowed to do. Write the boundaries down.