multi-agent

What problem does it solve?

This Skill addresses the critical limitations of single AI agents in handling complex, long-running tasks. It overcomes context limits, prevents constant interruptions by agents, and ensures high-quality outputs through structured review cycles. Engineers can delegate multi-hour tasks without constant supervision, freeing them to focus on higher-value work.

Core Features & Use Cases

• Autonomous Agent Teams: Orchestrates specialized agents (Explore, Plan, Architect, Dev, Test, Review, Scribe) to tackle complex projects.
• Signal-Based Decision Making: A central PM agent makes adaptive decisions based on real-time signals, eliminating hardcoded timeouts and ensuring continuous progress.
• Persistent Sessions & Remote Support: Agents maintain context across sessions and can operate on remote codebases via SSH, enabling seamless development workflows.
• Quality Assurance: Integrates mandatory review cycles to ensure iterative feedback and high-quality deliverables before task completion.
• Use Case: Assign the system to "investigate a device crash and create reproduction tests." Return hours later to find root cause analysis, deployed tests, and documented fixes, or a clear report on what context is needed to proceed.

Quick Start

To start a new task named 'my-feature' in your current project, first set up agents by running ./scripts/setup-agents . then initiate the task with ./scripts/task start my-feature --project .. You can check its progress using ./scripts/task status my-feature.