The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's architecture for fanning out and then aggregating worker results presents an indirect prompt injection surface.
Ingestion points: Data is ingested from terminal outputs and reports generated by parallel workers during the 'Aggregate' phase described in references/swarm-procedure.md.
Boundary markers: Instructions suggest formatting results into tables and issue one-liners to avoid 'raw worker dumps', but no technical boundary markers or 'ignore' instructions are mandated for the aggregation step.
Capability inventory: The skill coordinates worker agents and utilizes the worktree skill for isolated file operations.
Sanitization: There are no requirements to escape or validate worker-generated content before it is processed by the parent agent or included in the final report.

swarm