one-way-door
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided decision summaries to perform its classification task.
- Ingestion points: The required one-line summary of the pending decision and the optional decision ID from the decision registry.
- Boundary markers: Absent. The instructions do not provide delimiters or instructions for the agent to distinguish between the provided data and potential instructions embedded within the user summary.
- Capability inventory: The skill is restricted to decision classification and reporting. It lacks capabilities for arbitrary command execution, network requests, or persistent file system modifications beyond reporting its findings.
- Sanitization: No input validation or sanitization is defined for the decision summary strings before they are processed by the classification logic and regex patterns.
Audit Metadata