skills/simhacker/moollm/evaluator/Gen Agent Trust Hub

evaluator

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • [Prompt Injection] (SAFE): The skill uses 'blind evaluation' instructions to ensure the model ignores background context. This is a core architectural feature for objective scoring rather than an attempt to bypass safety guardrails.- [Data Exposure & Exfiltration] (SAFE): No hardcoded credentials or unauthorized network calls were found. File operations are restricted to reading inputs and writing evaluation reports within the defined workspace.- [Indirect Prompt Injection] (LOW): The skill processes untrusted committee outputs. 1. Ingestion points: read_file (output, rubric files). 2. Boundary markers: Absent; instructions do not specify delimiters for untrusted data. 3. Capability inventory: read_file, write_file (limited to local file system). 4. Sanitization: Absent; no explicit escaping or validation of external content. Analysis: Vulnerable surface exists but impact is limited by restricted tool access.- [Unverifiable Dependencies] (SAFE): No external package managers or remote code execution patterns were identified.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:39 PM