parallel-agents
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONNO_CODE
Full Analysis
- [Prompt Injection] (HIGH): The skill establishes an orchestration framework for agents to ingest and process untrusted external data (codebases).
- Ingestion Points: Files are read by agents like 'explorer-agent' and 'security-auditor' across orchestration patterns.
- Boundary Markers: The skill lacks delimiters to isolate untrusted code content from agent instructions, allowing embedded malicious instructions to hijack the orchestration logic.
- Capability Inventory: Orchestrated agents have high-privilege capabilities including multi-step code modifications, infrastructure deployment, and vulnerability exploitation.
- Sanitization: No input validation or sanitization is performed on data passed between agents or synthesized in reports.
Recommendations
- AI detected serious security threats
Audit Metadata