dispatching-parallel-agents

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (HIGH): The skill design facilitates Indirect Prompt Injection by instructing the agent to build a pipeline where untrusted data influences high-privilege actions.
  • Ingestion points: The 'Common Mistakes' and 'Agent Prompt Structure' sections in SKILL.md explicitly direct the agent to paste error messages and test names (external data) into sub-agent prompts.
  • Boundary markers: The skill mentions using 'Constraints', but these are provided as natural language instructions within the same context as the untrusted data, which is a weak defense against adversarial injections.
  • Capability inventory: The sub-agents created by this pattern are given broad authority to modify production source code ('Fix bugs in abort implementation') and run code within the environment to verify fixes.
  • Sanitization: The skill lacks any requirement for sanitizing, escaping, or validating the external content before it is interpolated into the prompts for the parallel agents.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 12:21 AM