Description-Behavior Mismatch
Medium
- Confidence
- 97% confidence
- Finding
- The manifest and description position the skill as a read/search interface, but the body instructs the agent to inspect and invoke arbitrary connector actions and discusses write/destructive operations. This creates a capability/expectation mismatch that can cause an agent or user to trust the skill as read-only when the underlying connector may expose state-changing actions, increasing the risk of unintended modification if the action list later expands or is misread.
