The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates data from parallel workers in Phase C ('Aggregate'). While reading output from subagents that may process untrusted data is a potential injection surface, the skill implements effective mitigations by instructing the agent to summarize results into one-line issue descriptions and compact tables instead of performing raw data passthrough.
[DYNAMIC_EXECUTION]: The skill describes orchestrating subagents and background processes ('launchers'). This is the intended primary purpose of the skill. It emphasizes isolation through specific worktree and output directory assignments to ensure that parallel tasks do not interfere with each other or the host environment.

swarm