The Agent Skills Directory

[SAFE]: The skill implements a formal trust and integrity system for multi-agent communication. It uses JSON schemas (context-envelope.schema.json) to validate the origin, sensitivity, and integrity (via SHA-256 digests) of shared context.
[COMMAND_EXECUTION]: Orchestration is performed using native platform primitives (spawn_agent, list_agents, interrupt_agent). The skill enforces strict naming conventions and delegation limits, such as the 'one-active-grandchild' limit, to maintain control over the task graph.
[DATA_EXFILTRATION]: No unauthorized network operations or sensitive file access patterns were identified. The skill explicitly limits task messaging to the current root task tree and requires operator approval for any external, destructive, or credentialed writes.
[EXTERNAL_DOWNLOADS]: References to external schemas (json-schema.org) and vendor-owned documentation sites (xsyetopz.github.io) are used for structural validation and policy reference, which is standard for this framework.
[PROMPT_INJECTION]: The skill uses behavioral role templates and explicit ownership boundaries to scope sub-agent behavior. It includes robust error-handling logic that escalates reasoning effort in response to failures rather than bypassing established safety constraints.

fourth-wall