santa-method
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill framework involves processing data generated by an AI agent (generator), creating a standard surface for indirect prompt injection where the review agent could be influenced by malicious content in the generated output.
- Ingestion points: The
REVIEWER_PROMPTdefined inSKILL.mdincorporates the{output}variable from the generator agent. - Boundary markers: The framework uses Markdown headings (e.g.,
## Output Under Review) as delimiters to provide structure and context for the agent. - Capability inventory: Reviewer agents generate structured JSON verdicts, and the
fix_agentis instructed to modify the original output to address critical issues. - Sanitization: The process relies on dual independent reviews and a strict rubric-based evaluation to identify and filter out anomalous or adversarial content.
Audit Metadata