advanced-evaluation
Pass
Audited by Gen Agent Trust Hub on Apr 30, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill's primary functionality creates an attack surface for Indirect Prompt Injection.
- Ingestion points: Untrusted data enters the agent context through the
{prompt},{response},{response_a}, and{response_b}placeholders in the evaluation prompt templates found inSKILL.md. - Boundary markers: The templates utilize Markdown headers (e.g.,
## Original Prompt,## Response to Evaluate) to delimit untrusted content from the evaluation instructions. - Capability inventory: The skill facilitates LLM text generation based on these templates, as described in
references/implementation-patterns.mdand demonstrated inscripts/evaluation_example.py. - Sanitization: There is no evidence of input sanitization or explicit "ignore embedded instructions" warnings within the templates to further mitigate adversarial input.
Audit Metadata