dispatch-parallel
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides architectural instructions for managing complex multi-agent workflows. It focuses on reliability and isolation rather than introducing dangerous behaviors.
- [PROMPT_INJECTION]: The skill handles untrusted user input by wrapping it in structured XML delimiters (e.g., , ) and providing rigid prompt templates for sub-agents. This structure reduces the likelihood of sub-agents being diverted from their specific tasks.
- Ingestion points: User-provided task descriptions are ingested via the $ARGUMENTS variable.
- Boundary markers: The orchestrator uses specific tags and prefixes (Zero-shot CoT) to isolate the task from the agent's reasoning.
- Capability inventory: The skill utilizes the Task tool to delegate work, but enforces strict file ownership to prevent cross-contamination.
- Sanitization: The skill uses a 'Self-critique' phase for sub-agents to verify their own output against the rubric.
- [DATA_EXFILTRATION]: No network activity or sensitive data access patterns were found. The skill emphasizes local filesystem isolation and auditability through state storage in .shannon/state/.
- [COMMAND_EXECUTION]: The skill orchestrates parallel execution via tool calls but includes a mandatory 'Independence validation' step to identify and prevent potential race conditions or file conflicts.
Audit Metadata