The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes an architecture where a root agent ingests data from multiple subagents through system URIs.
Ingestion points: Data is received from subagents via agent://<id> and history://<id> references, as well as metadata returned from the task tool.
Boundary markers: The instructions mandate structuring shared context with Goal, Constraints, and Contract to delimit assignments.
Capability inventory: The root coordinator is instructed to integrate changes, which may involve file writes and running validation suites or user paths to verify the head state.
Sanitization: The skill explicitly instructs the agent to "Inspect each claimed artifact and reconcile conflicts against the declared ownership and contracts" and to check if results are "incomplete or suspicious," providing a manual review step for external content.

orchestrate-omp