The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates agent orchestration based on potentially untrusted external inputs, creating a surface for indirect instructions.\n- Ingestion points: External task descriptions provided via the --task argument in npx agentic-flow hooks task-orchestrate and the goal field in swarm.autoOrchestrate in SKILL.md.\n- Boundary markers: The skill lacks explicit instructions or delimiters to help the agent differentiate between orchestration instructions and data content within the task descriptions.\n- Capability inventory: The orchestration system allows spawning agents with significant capabilities, such as coder and tester (mentioned in SKILL.md), which may execute code or modify the local environment.\n- Sanitization: There is no mention or implementation of validation or sanitization for the task goals before they are processed by the swarm.

Swarm Orchestration