The Agent Skills Directory

[Task Delegation and Orchestration]: The skill provides instructions for delegating tasks to subagents. While this involves processing natural language instructions which can be a vector for indirect prompt injection, the skill centers on orchestration logic within the author's established ecosystem.
[Security Best Practice Integration]: It strongly advocates for and provides examples of Human-in-the-Loop (HITL) workflows. By configuring interrupt_on for sensitive tools like execute_sql or write_file, the skill helps users implement a robust security boundary for autonomous operations.
[Data Ingestion and Handling]: Tools like task and write_todos (defined in SKILL.md) ingest external data into the agent context. While the skill does not detail automated sanitization, it emphasizes manual boundary markers via HITL interrupts. This ensures that capabilities such as SQL execution, file writes, or deployments are manually validated by a human, effectively mitigating potential risks from instructions embedded in processed data.

deep-agents-orchestration