santa-method
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a framework where reviewer subagents are provided with potentially untrusted output and task specifications, creating a surface where malicious input could attempt to override reviewer instructions.
- Ingestion points: The
REVIEWER_PROMPTtemplate inSKILL.mddirectly interpolates{task_spec},{output}, and{rubric}variables into the subagent's context. - Boundary markers: The prompt uses Markdown headers as structural delimiters, but lacks explicit instructions for subagents to ignore or treat embedded instructions as data only.
- Capability inventory: The framework utilizes the
Agenttool to spawn subagents, which may have access to a variety of system tools depending on the platform's execution environment. - Sanitization: The documentation does not specify methods for escaping or validating variable content before interpolation into subagent instructions.
Audit Metadata