The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages background tasks and local script execution to maintain a stateful orchestration loop.
Evidence: Utilizes node handoff.mjs and background bash scripts (wait-for-ready.sh) to monitor task completion without polling.
[REMOTE_CODE_EXECUTION]: Orchestrates the launch of sub-agents with permission gates disabled, allowing for broad local access and execution capabilities.
Evidence: Explicitly instructs the agent to use high-privilege flags such as --dangerously-bypass-approvals-and-sandbox and --permission-mode auto for builder harnesses.
[INDIRECT_PROMPT_INJECTION]: Ingests and acts upon data generated by secondary agents, presenting a surface for state manipulation if a builder is compromised.
Ingestion points: Reads progress from board output and detailed status from builder handoff documents.
Boundary markers: Separates orchestrator state (ledger) from builder communication (handoff docs).
Capability inventory: Can write files, manage background processes, and control multiple agent panes.
Sanitization: Employs structured CLI tools for file modifications to prevent manual parsing errors, but lacks natural language sanitization for builder reports.

complete