Intent-Code Divergence
- Category
- Not specified by scanner
- Confidence
- 93% confidence
- Finding
The skill claims command execution is restricted by a whitelist, but the document elsewhere instructs the agent to create and run local shell scripts, Python commands, and fetch jobs without defining any actual enforcement boundary. This mismatch is dangerous because users or downstream agents may trust the safety claim and execute generated commands that can read local secrets, invoke network calls, or modify files.
- Content
