The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines multiple patterns (Swarm, Hierarchical, Generator/Critic, Jury) that ingest untrusted external data which is then processed by autonomous agents.
Ingestion points: The task_description variable is used as the primary input for the Swarm, Hierarchical, and Generator/Critic execution methods. The question variable is used as the input for the Jury decision method.
Boundary markers: Absent. The provided code templates do not include clear delimiters (e.g., XML tags or triple quotes) or specific system instructions to the spawned agents to ignore potential commands embedded within the task descriptions.
Capability inventory: While no direct shell or network operations are present in the skill's markdown, the patterns themselves represent high-level agentic capabilities (coordination, implementation, validation) that would be executed by the underlying framework.
Sanitization: Absent. The skill does not provide logic for validating, escaping, or sanitizing the input variables before they are passed to the agent execution methods.

agentica-spawn