fnl-advisor
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill takes user input from the $ARGUMENTS variable and file content from the local environment and passes them to an external 'advisor' model. This creates a surface for indirect prompt injection where malicious content in those inputs could influence the advisor's reasoning or cause it to attempt to invoke other skills.\n
- Ingestion points: User input via the $ARGUMENTS parameter and contents of files referenced in the prompt.\n
- Boundary markers: The briefing explicitly ends with an instruction to the advisor: "This is analysis only. Do NOT edit, create, or delete any files. Do NOT write code."\n
- Capability inventory: The skill triggers external model calls and allows the advisor to be instructed to invoke other skills (via fnl-delegate).\n
- Sanitization: No sanitization or filtering is performed on the user-provided arguments or file content before they are forwarded to the advisor model.
Audit Metadata