The Agent Skills Directory

[SAFE]: Multi-Agent Orchestration and Lifecycle. The skill defines a structured 7-phase lifecycle (ASSESS through CLEANUP) for the creation and management of parallel Claude agent teams. It enforces safe initialization with TeamCreate and ensures proper cleanup with shutdown_request and TeamDelete to prevent orphaned processes or resource leaks.
[SAFE]: Resource Isolation and Integrity. The skill uses a dedicated file-ownership-protocol.md to manage file system access. By requiring teammates to have exclusive_write permissions and validating these against a central ownership_map, the system prevents conflicting writes and unauthorized modifications during parallel execution.
[SAFE]: Communication Security. The orchestrator adheres to a hub-spoke model where all sub-agent communication is routed through the lead agent. This architecture, combined with the use of predefined messaging templates and manual approval modes (e.g., plan mode), ensures that sub-agents operate within defined constraints and allows for human-in-the-loop checkpoints for high-risk tasks.
[SAFE]: External Interactions. All external tool calls are directed to the trusted Claude Code Agent Teams API. There are no external dependency installations or remote code execution patterns involving untrusted third-party sources.

Rally