The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a pattern for merging results from subagents, which identifies a potential surface for indirect prompt injection where a child agent's output could influence the parent's logic.
Ingestion points: Merging results from child agents via the harness.spawn mechanism documented in SKILL.md.
Boundary markers: The skill strongly recommends the use of structured output_schema and strict return shapes to delimit subagent output.
Capability inventory: Parent agents are instructed to delegate tasks to subagents which can run in isolated isolation modes (Worktree, Sandbox) to limit environmental impact.
Sanitization: Recommends performing validation, reconciliation, and schema checks during the result merge phase to handle format drift or conflicting outputs.

subagent-orchestration