The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a central coordinator ("Queen") that decomposes user objectives into sub-tasks for delegation to specialized worker agents (Research, Code, Analyst, Test). This design creates an indirect prompt injection surface where instructions embedded within the processed tasks could attempt to manipulate the coordinator's logic or influence the behavior of the spawned workers.
Ingestion points: The skill processes user tasks via the $TASK variable in its lifecycle hooks and through the task_orchestrate command string.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" protocols for handling external task data.
Capability inventory: The skill possesses significant capabilities, including agent spawning (agent_spawn), task delegation, and shared memory management across the swarm.
Sanitization: No explicit validation, filtering, or sanitization of input task descriptions is observed in the coordinator's workflow.
[COMMAND_EXECUTION]: The skill utilizes shell commands within its pre and post hooks to initialize swarm topology and generate reports.
Evidence: Execution of echo, date, and jq alongside platform-specific mcp__claude-flow__ tools to manage state and monitor performance.

agent-hierarchical-coordinator