llm-judge-creator
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest external data from the workspace and user-provided reports to generate evaluation prompts.
- Ingestion points: The agent is instructed to read project files (e.g.,
CLAUDE.md,product-marketing-context.md) and user-provided "issue reports" or "annotated traces" from the workspace. - Boundary markers: The prompt template uses
{{INPUT}}and{{OUTPUT}}placeholders, but the skill does not provide explicit boundary markers or instructions to ignore malicious commands embedded within the failure examples or project context. - Capability inventory: The skill is entirely instructional (Markdown) and does not specify tools for file writing, network access, or command execution. It relies on the agent's base capabilities.
- Sanitization: No sanitization or validation logic is defined for the ingested data.
Audit Metadata