santa-method
Pass
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a structural framework for quality assurance and verification. It contains no executable scripts, remote downloads, persistence mechanisms, or credential-harvesting patterns. It explicitly includes security-focused checks (secrets detection and injection prevention) within its recommended evaluation rubrics.\n- [PROMPT_INJECTION]: The skill architecture creates a surface for indirect prompt injection by design, as it processes the output of a primary generator agent as input for reviewer subagents.\n
- Ingestion points: The
REVIEWER_PROMPTinSKILL.mdinterpolates the{output}and{task_spec}variables into a new agent context.\n - Boundary markers: The instructions use Markdown headers (e.g.,
## Task Specification) to delimit instructions from processed data, providing some degree of structural separation.\n - Capability inventory: The skill uses the
Agenttool to spawn subagents which then process the interpolated content.\n - Sanitization: No explicit sanitization or escaping of the interpolated generator output is performed. This is mitigated by the persona instructions in the reviewer prompt which mandate rigorous adherence to the rubric.
Audit Metadata