simmer-judge-board
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze potentially untrusted data such as candidate artifacts and evaluator output. This creates a surface for indirect prompt injection where malicious instructions embedded in the data could influence the judging or deliberation process. 1. Ingestion points: artifact candidate text, workspace files, and evaluator stdout/stderr. 2. Boundary markers: the prompt templates use textual headers to guide the agent, but lack strict data delimiters or robust sanitization for the untrusted inputs. 3. Capability inventory: the skill includes file reading, peer skill invocation (simmer-judge), and tool-based search functionality. 4. Sanitization: no explicit sanitization or validation mechanisms are defined for the processed content.
Audit Metadata