What problem does it solve? Coordinating multiple AI coding agents on a single task is opaque and error-prone: agents pause mid-task waiting for input, terminal output gets truncated, and users lose visibility into what each step is doing. This Skill chains scout, planner, worker, and reviewer agents with file-based handoff so every step runs in an observable overlay the user can watch or take over. ## Core Features & Use Cases - Four-role agent chain: Runs Scout (codebase recon), Planner (implementation plan), Worker (code changes), and Reviewer (validation and fixes) using different CLI agents like cursor-agent, pi, and codex. - File-based handoff: Passes structured artifacts (context.md, plan.md, impl.md) through a shared chain directory, with a single progress.md log every agent appends to. - Auto-continue handling: Detects when agents pause with prompts like "Should I continue?" and automatically sends input to keep them working, while respecting user takeover. - Use Case: Ask pi to implement a feature; it scouts the codebase for relevant code, generates a plan, implements all tasks with auto-continue, then reviews and fixes issues, leaving a complete progress log. ## Quick Start Ask pi to run a foreground chain that scouts the codebase, plans, implements, and reviews a feature you describe, then read the progress.md file it produces.
