The Agent Skills Directory

[SAFE]: The skill describes a complex but secure orchestration pattern for sub-agents. It uses a coordination worktree and positional signaling (cd into a specific directory) to trigger custom hooks (doctrine worktree create-fork) that isolate workers.
[COMMAND_EXECUTION]: The skill utilizes standard git commands (git status, git rev-parse, git merge-base) and project-specific CLI tools (doctrine, dispatch). These are used for environment verification and state management, consistent with the skill's purpose as an agent orchestrator.
[REMOTE_CODE_EXECUTION]: While the skill involves spawning sub-agents (via the Agent tool), it documents rigorous isolation measures. The use of PreToolUse hooks to wrap shell execution in a bwrap (Bubblewrap) jail indicates a high level of security awareness regarding sub-agent confinement.
[PROMPT_INJECTION]: The skill includes instructions for creating worker prompts, but these are focused on safety and integrity (e.g., the 'BASE GUARD' block to prevent unauthorized modifications) rather than bypassing safety filters.
[DATA_EXFILTRATION]: No network operations or suspicious data access patterns were identified. Data movement is confined to local git worktree operations and gated MCP tools.

dispatch-agent