evaluator

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documentation explicitly instructs the agent to use 'Bash execution' for testing CLI tools and 'evaluate_script' for backend verification. This capability allows the agent to execute shell commands based on the context of the task being evaluated.
  • [DATA_EXFILTRATION]: The skill suggests using 'curl' or 'httpie' for response assertion during API and backend testing. This enables the agent to initiate external network requests to verify functionality.
  • [PROMPT_INJECTION]: The 'Evaluator Agent Template' creates a surface for indirect prompt injection by interpolating untrusted content directly into the agent's instructions.
  • Ingestion points: The agent reads 'git diff' output and 'acceptance criteria' from bead metadata, both of which can contain externally influenced or malicious content.
  • Boundary markers: Absent; the prompt template uses placeholders without delimiters or instructions for the agent to treat the data as untrusted or to ignore embedded instructions.
  • Capability inventory: Access to bash command execution, network utilities (curl/httpie), and browser automation via 'dm-work:browser-qa'.
  • Sanitization: There is no evidence of validation, escaping, or sanitization of the injected criteria or diff data before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 03:50 PM