01Configure agents with varying sandbox modes (read-only, workspace-write, danger-full-access) and reasoning effort levels (minimal to xhigh).
02Spawn multiple specialized AI agents with distinct configurations (model, sandbox, effort, personality).
03Generate agent configurations interactively using an 'Agent Factory' wizard tool.
04Ensures security with default read-only sandboxes and enforcement of security overrides that can only tighten, not loosen, permissions.
054 GitHub stars
06Override server-level agent parameters with per-invocation settings for prompts, working directories, and security policies.
