dispatching-parallel-agents
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests external data into prompts for sub-agents. \n
- Ingestion points: External test failure logs and error messages are used to populate sub-agent tasks as shown in the 'Agent Prompt Structure' section. \n
- Boundary markers: The skill suggests providing 'Constraints' and 'Clear goals', but does not mention technical delimiters or sanitization to isolate untrusted error log data from sub-agent instructions. \n
- Capability inventory: The workflow uses a 'Task()' primitive to spawn sub-agents which may have the ability to read and modify the filesystem. \n
- Sanitization: There is no guidance on sanitizing or escaping ingested data before passing it to sub-agents. \n- [DYNAMIC_EXECUTION]: The skill relies on the 'Task()' primitive to dynamically create and execute new agent sessions. This represents a dynamic execution surface where instructions in the skill lead to the creation of new computational contexts.
Audit Metadata