cognitive-baseline-eval
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWNO_CODEPROMPT_INJECTION
Full Analysis
- [NO_CODE] (SAFE): The skill is entirely composed of Markdown instructions. It contains no Python, JavaScript, or shell scripts, and no external dependencies are declared.
- [PROMPT_INJECTION] (LOW): The skill involves processing 'adversarial prompts' and 'transcripts' to evaluate AI behavior. This presents an indirect prompt injection risk if the external content contains malicious instructions designed to subvert the evaluation logic.
- Ingestion points: External data referred to as 'transcript' and 'standardized adversarial prompts'.
- Boundary markers: None provided in the skill body to isolate external content.
- Capability inventory: The skill is limited to internal reasoning and tier-based scoring; it lacks the capability to write files, execute system commands, or make network requests.
- Sanitization: No logic is present to sanitize or filter potentially malicious instructions within the evaluated content.
Audit Metadata