dispatching-parallel-agents
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where an agent ingests untrusted data and uses it to generate prompts for other agents, creating a potential injection surface.
- Ingestion points: The agent is instructed to read external data such as test failure logs (e.g., error messages like 'interrupted at') and source code files (e.g.,
src/agents/agent-tool-abort.test.ts) to define tasks. - Boundary markers: The skill lacks guidance on using delimiters or explicit 'ignore embedded instructions' markers when interpolating this external content into the descriptions for
Task()calls. - Capability inventory: The skill leverages the
Task()function to spawn new agents, which may inherit broad tool access or capabilities. - Sanitization: There are no instructions provided to sanitize, escape, or validate the content retrieved from error logs or external files before it is passed to the sub-agent, allowing malicious strings in logs to potentially influence the sub-agent's behavior.
Audit Metadata