Security Boundaries
What PermitGraph does and does not execute.
What this is
Security boundaries define what the scanner is allowed to inspect.
When to use it
Use this page before scanning untrusted repositories or sharing generated artifacts.
Command or example
uv run agent-permit scan . --ciOutput to expect
The scanner reads files and configuration. It does not run agent code, MCP servers, CI workflows, package scripts, or external tools.
How to interpret it
PermitGraph is intentionally static in the MVP path. This keeps evidence repeatable and avoids giving the scanner the same dangerous access it is reviewing.
Common mistakes
- Expecting the scanner to prove runtime behavior.
- Adding destructive tools to the scan path.
- Publishing generated evidence without review.