Tp4
- Category
- MCP Tool Poisoning
- Confidence
- 99% confidence
- Finding
The declared description presents an end-user assistant workflow for handling several real-world support/operations tasks via AI and authorized tools. The supplied code does none of that. It only validates a saved result file against a contract and local artifact hashes, explicitly stating it does not authenticate, call APIs, or execute actions. This is a materially different primary purpose, not just an internal helper detail for the described skill. No relevant assistant behaviors, user-facing task handling, reminders, follow-up flows, or external tool interactions are implemented in this code chunk.
- Content
