The five rungs. Every AI capability on AgentMaps carries a verification level: Listed → Info checked → Risk reviewed → Install-verified → Interface-verified. Each rung is stricter than the last. Listed means the capability exists in the catalog and nothing more. Info checked adds source, docs, and metadata confirmation. Risk reviewed adds a hands-on read of permissions and risks from the source repo. Install-verified adds a real install-path check. Interface-verified adds a check of the tool or API surface.
Why a ladder, not a flag. A single yes/no "verified" flag hides how much was actually done. The ladder is honest: you can see at a glance whether the entry was skimmed or exercised, and whether the level of check matches the risk the capability carries.
What a high level does not mean. A high verification level is not an endorsement for production use. It tells you how far the catalog team looked, not that the capability is safe or effective in your setup. Always combine verification level with the risk level and your own trial in an isolated environment before adopting.