quality-auto-test

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through its ingestion of repository metadata and documentation. \n
  • Ingestion points: File contents from requirement documents (REQ-*.md) and source code are used to formulate test scenarios during the S_SOURCE and S_INFRA phases. \n
  • Boundary markers: Project data is interpolated into sub-agent instructions without the use of dedicated delimiters (e.g., XML tags). \n
  • Capability inventory: The orchestrator is authorized to use shell execution (Bash), file modification (Write/Edit), and agent spawning (spawn_agents_on_csv). \n
  • Sanitization: The skill relies on natural language constraints within the system instructions to prevent sub-agents from modifying the primary source code or executing recursive calls.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 11:26 PM
Security Audit — agent-trust-hub — quality-auto-test