eval-guidance-actionability

Pass

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface.
  • Ingestion points: The skill ingests 'assistant response text' in SKILL.md for evaluation.
  • Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' instructions for the untrusted input.
  • Capability inventory: None. The skill has no access to shell, file system, or network tools; it only generates JSON text.
  • Sanitization: No sanitization or validation of the input text is implemented.
  • Note: While an attacker could potentially manipulate the score or rationale in the resulting JSON, the lack of tool-use capabilities prevents any high-severity impact.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 19, 2026, 10:32 AM