The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages a multi-agent workflow using background shell scripts (wait-for-ready.sh) and terminal multiplexers like tmux and herdr. It monitors system processes to manage the lifecycle of these background tasks.
[PRIVILEGE_ESCALATION]: The orchestrator is designed to launch sub-agents ("builders") with relaxed security constraints, such as disabling sandboxes via flags like --dangerously-bypass-approvals-and-sandbox. The instructions mandate that the agent must explicitly ask the user for confirmation before enabling this autonomy posture.
[INDIRECT_PROMPT_INJECTION]: The skill ingests status reports and disagreements from files ("handoff docs") written by other agents. This creates an attack surface where a compromised sub-agent could attempt to influence the orchestrator. Mitigation strategies included in the instructions involve using specific CLI tools for file interaction and a "council" mechanism for resolving judgment calls.
[COMMAND_EXECUTION]: The provided wait-for-ready.sh script scans the system's process list (ps) to identify and terminate stale instances of itself, ensuring efficient resource management during long-running tasks.

complete