Tp4
- Category
- MCP Tool Poisoning
- Confidence
- 96% confidence
- Finding
This is another true documentation-to-behavior mismatch affecting the core purpose of the skill. If the shipped code chunk is only cleanup/vetting and not export functionality, users and orchestrators may grant powerful permissions and run the skill under false assumptions about what it can safely do, which is dangerous for a tool handling sensitive ChatGPT history.
- Content
