dispatching-parallel-agents
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs agents to analyze and modify external source code files (e.g., test files). This creates an attack surface where malicious content or instructions embedded in those files could potentially influence the sub-agent's actions.
- Ingestion points: Sub-agents are assigned to process local files such as
src/agents/agent-tool-abort.test.tsorbatch-completion-behavior.test.tsin the SKILL.md file. - Boundary markers: The instructions advocate for "isolated context" and state that sub-agents "should never inherit your session's context or history," which helps mitigate cross-task contamination.
- Capability inventory: The delegated agents have the ability to read code, identify root causes, and modify files to resolve failures.
- Sanitization: The skill does not provide specific guidance on sanitizing or validating the content of the files before the agent processes them, relying instead on the isolation of the agent's environment.
Audit Metadata