alternative-solution-designer
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided problem descriptions, conversation history, and external files to analyze and reframe problems. This architecture presents a surface for indirect prompt injection where malicious instructions embedded in the processed data could attempt to influence agent behavior.
- Ingestion points: User prompts, conversation history, and local files (SKILL.md Step 1).
- Boundary markers: There are no instructions to use specific delimiters or ignore embedded directives when processing this data.
- Capability inventory: The skill utilizes network search tools to verify technical concepts mentioned in the user input (Step 0).
- Sanitization: The instructions do not specify any validation or sanitization of user data before it is processed or used in tool calls.
Audit Metadata