If you already use GPT or Claude and you are trying to connect a few tools just to save time, this is the mistake to avoid: treating Claude like just another tool and assuming the highest score means the best fit. The cost shows up when you search in a browser, jump back to chat to restate context, then return to your editor again. AI tools are starting to compete not just for coding work, but for those little fragments of time you lose switching back and forth.
The first page of a plugin README should be the trust boundary. Before features, a normal user needs four answers: what the plugin can access, what login it needs, what extra dependencies it relies on, and what can change later. Many people think they need a stronger model. What they actually need is fewer windows to bounce between.
The repo anthropics / claude-plugins-official makes that boundary explicit. Its README says you should trust a plugin before installing, updating, or using it, and it warns that Anthropic does not control the MCP, files, or other software inside a plugin and cannot verify that they will behave as expected or stay unchanged [S001]. The FAQ on claude.com/plugins adds an important limit: submissions get basic automated review, and Verified means extra quality and safety review, not a blanket guarantee [S002].
The practical reason this matters is simple. One GitHub plugin listing in the official marketplace at one point showed only plugin.json and .mcp.json, required GITHUB_PERSONAL_ACCESS_TOKEN, and did not include a README or authentication guide [S003]. When the trust boundary is missing, users do not just get thinner docs. They get extra guessing, extra tab switching, and a fast path to thinking the plugin is broken.
So the next time a plugin promises convenience, do not start with the feature list. Start with the README. If it does not clearly spell out permissions, dependencies, login steps, and risk, treat that as product information, not a minor docs gap. Share this with the person who keeps trying plugins to save time and ends up doing more manual setup instead.