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.mdoragents/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