surrogate-verifier
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external, potentially untrusted content, including skill definition files (SKILL.md) and task prompts.
- Ingestion points: Input artifacts such as skill definitions, task prompts, and previous outputs are processed during Phase 1 (Skill Analysis) and Mode 2 (Failure Diagnostics).
- Boundary markers: The Information Isolation Protocol advises running the verifier in a separate agent spawn or fresh session to prevent context leakage, but the skill does not define specific data-level delimiters or 'ignore' instructions for the ingested text.
- Capability inventory: The skill workflow references external scripts (scripts/eval_assertions.py and scripts/run_evals.py) and assertions verify usage of tools like Read, Write, and WebSearch.
- Sanitization: The instructions do not describe methods for sanitizing or escaping the content being analyzed to prevent embedded prompts from being interpreted by the LLM.
Audit Metadata