agent-evaluation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, command executions, or network exfiltrations were detected. The skill is entirely instructional and serves as a meta-framework for quality assurance.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a 'Judge Prompt Template' which acts as a surface for ingesting external data (agent outputs and original tasks). While this represents a potential injection vector if the data being evaluated contains malicious instructions, the risk is mitigated by the structured nature of the evaluation rubric provided.
- Ingestion points: Placeholder fields in the 'Judge Prompt Template' within SKILL.md.
- Boundary markers: Absent; the template uses square bracket placeholders but does not include explicit delimiters or instructions to ignore embedded commands within the content being judged.
- Capability inventory: None; the skill does not contain any scripts, tool calls, or network capabilities.
- Sanitization: Absent; the skill does not define methods for filtering or escaping the text being evaluated.
Audit Metadata