If you already use GPT or Claude and want to stitch tools together to save time, this is the trap: you think you need a stronger model, but the real cost is window switching. A plugin README should define the trust boundary on page one, not hide it behind features. That is the more useful reading of anthropics / claude-plugins-official [C001].
A lot of people treat Claude like just another scorecard: pick the stronger model and move on. I think that misses the real pain. AI tools are starting to compete on the minutes you lose bouncing between browser, chat, and editor. Many people think they need a stronger model. What they really need is fewer windows.
That is why the first page matters. An official store is not an official guarantee. Anthropic's public repo says trust comes before install, update, and use. If trust comes first, then the README is not marketing copy. It is the boundary line.
Issue #285 is a good example of the failure mode. A GitHub plugin showed plugin.json and .mcp.json, asked for a GITHUB_PERSONAL_ACCESS_TOKEN, and had no plain-English setup guide. A normal user does not read that as “powerful.” They read it as confusing, broken, or easy to misjudge.
The FAQ adds the missing limit: submission gets basic automated review, and Verified means extra checks, not a blanket promise. So the practical rule is simple: read permissions, dependencies, auth steps, and risk notes before you read features.
If that framing helps, share it with the next person who thinks “official store” automatically means “safe enough.”