start

Orchestrates the full agentic coding flywheel from goal intake through implementation and review.

3|1|Updated Apr 15, 2026
One-click install
npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill start-burningportra
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: start
Source: https://github.com/burningportra/agent-flywheel-plugin/tree/main/skills/start
Command: npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill start-burningportra

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coding agents stall mid-task, lose context across sessions, skip review passes, and collide when multiple agents work in parallel. This Skill drives a gated, resumable multi-agent workflow so coding work progresses through scan, discover, plan, implement, and review phases with checkpoints and user approval at every seam. ## Core Features & Use Cases - Full flywheel orchestration: Runs the complete scan → discover → plan → implement → review loop with atomic checkpoints stored on disk for cross-session resume. - Multi-agent swarm coordination: Spawns model-diversified agent lanes (Claude, Codex, Gemini) via NTM tmux panes with Agent Mail for file reservations, progress messaging, and conflict prevention. - Gated decision flow: Every user decision routes through structured AskUserQuestion menus — goal framing, plan approval, bead quality scoring, launch confirmation, and wrap-up. - Use Case: Point it at a repository with a goal like "add OAuth login", and it profiles the repo, generates a plan, creates dependency-ordered task beads, dispatches parallel implementation agents, and runs adversarial review before wrap-up. ## Quick Start Run /agent-flywheel:start with an optional goal sentence to launch the full flywheel workflow on the current project.

Frequently Asked Questions about start

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I start a multi-agent coding workflow in Claude Code?

Run /agent-flywheel:start with an optional goal. The skill detects existing sessions and beads, shows a state-aware menu, then drives scan, discover, plan, implement, and review phases with checkpoints after every step.

How does the flywheel resume work after a session ends?

State is persisted to .pi-flywheel/checkpoint.json after every step. On restart, the skill reads the checkpoint, runs a drift check, and offers resume options including an auto-swarm in-flight resume mode.

Can I run parallel coding agents without merge conflicts?

Yes. The skill uses NTM tmux panes plus Agent Mail file reservations so agents claim files before editing. A pre-commit guard and shared-write hotspot analysis detect contention before launch and can force coordinator-serial mode.

What happens if NTM or Agent Mail is not installed?

The skill degrades gracefully: it detects missing tools in preflight, warns in the banner, and falls back to single-agent execution via the Agent tool. Doctor checks surface remediation steps through /agent-flywheel:flywheel-setup.

Why does the flywheel ask so many confirmation questions?

Every user decision must go through explicit AskUserQuestion gates by design — plan approval, bead quality scores below 0.75, launch confirmation, and recovery branches. This prevents agents from making implicit choices on destructive or high-cost actions.