The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents workflows where a 'coordinator' agent synthesizes artifacts, findings, and evidence from multiple 'worker' agents. This multi-agent delegation pattern creates a surface for indirect prompt injection, as the coordinator must process potentially unvetted natural language or code outputs generated by other agents. If a worker agent is compromised or processes untrusted external data, its output could contain instructions designed to manipulate the coordinator's logic during the 'Synthesis protocol' phase.

parallel-agents