orchestrate-engineering
Pass
Audited by Gen Agent Trust Hub on Aug 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external and potentially untrusted data, such as task acceptance criteria and system logs, to drive orchestration and delegation decisions.
- Ingestion points: Untrusted data enters the context when the agent reads the "task’s acceptance criteria" (SKILL.md) and analyzes "Large test or build output" (references/context-budget.md).
- Boundary markers: The instructions mandate providing workers with the "minimum task-local context" and a "Bounded scope" (SKILL.md), which serves to isolate the main context from potential injection in subtasks.
- Capability inventory: The skill manages subagents that are granted specific "Allowed actions and permissions," which may include file system access or command execution.
- Sanitization: The skill includes a dedicated "Integrate and verify" phase requiring the agent to "Validate worker claims against source files, commands, or authoritative documentation" and "Deduplicate findings and resolve contradictions" before accepting results.
Audit Metadata