eval-accuracy

Pass

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • Prompt Injection (LOW): The skill processes untrusted assistant responses, creating a surface for indirect prompt injection where malicious input could influence the evaluation results.\n
  • Ingestion points: SKILL.md (assistant response text input requirement).\n
  • Boundary markers: Absent; no delimiters or instructions to ignore embedded commands are present.\n
  • Capability inventory: None identified across SKILL.md or agents/openai.yaml; the skill lacks shell, network, or file-system access.\n
  • Sanitization: Absent.\n- No Code (SAFE): The skill is entirely prompt-based and does not include Python scripts, Node.js packages, or binary executables.\n- Data Exfiltration (SAFE): No mechanisms for network communication or sensitive data access were found.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 19, 2026, 10:33 AM