The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses platform-native tools, such as the Agent tool (Claude Code) or spawn_agent (Codex), to orchestrate task execution in isolated sub-agents and dynamically generates instructional briefs to manage their context.
[PROMPT_INJECTION]: The protocol ingests user-provided task descriptions and external skill metadata to make routing decisions, which represents an attack surface for indirect prompt injection, though this is mitigated by predefined venue heuristics.
[SAFE]: Implements a mandatory depth-guard mechanism (the COP-RESOLVED marker with depth: 1) to prevent uncontrolled recursive sub-agent spawning, ensuring the agent remains within stable operational bounds.
[SAFE]: Requires explicit human approval via the /ask-gate mechanism for any proposed self-improvement updates to its own execution logic, preventing unvetted autonomous modifications.

caller-options