What it covers. Permission scope is the honest per-capability answer to "what can this actually touch?" AgentMaps tracks seven axes: shell execution, file write, database access, browser control, local file read, external network access, and sensitive-token handling. Each capability's detail page shows which axes it needs and which stay off.
Why per-axis. A single "trusted" flag lumps very different risks together. A tool that only writes to one temp directory is not the same as one that runs arbitrary shell commands, even if both technically "touch your system." The seven axes let you make one decision per axis rather than one for the whole tool.
How to use it. When you evaluate a capability, map its permission scope against what the task actually needs. A summarizer that asks for shell access is a red flag — the permission surface exceeds the job description. Prefer capabilities whose scope matches the work; if the scope is wider than the job, look for a narrower alternative or scope the credentials tightly.