flywheel-duel

Runs state-aware multi-agent adversarial duels and routes artifacts into flywheel pipelines.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running multi-agent adversarial duels inside a structured coding workflow requires manual coordination: checking agent availability, picking the right duel mode for the current phase, and moving the resulting reports into the correct pipeline. This Skill automates that orchestration so duel outputs flow directly into discovery, planning, or review without manual file shuffling. ## Core Features & Use Cases - Pre-flight gating: Verifies ntm presence and at least two healthy agent CLIs (claude, codex, gemini) before starting, surfacing blockers through an interactive prompt. - Phase-aware mode detection: Reads the flywheel checkpoint to auto-select duel mode (ideas, architecture, reliability), output path, and top-N defaults based on the current workflow phase. - Artifact routing and chaining: Moves duel reports into docs/discovery, docs/plans, or docs/reviews and chains into flywheel_discover, flywheel_plan, or per-bead review with provenance metadata. - Use Case: During the planning phase, invoke the duel to have multiple agents cross-score architecture proposals, then automatically feed the consensus plan into the plan alignment check. ## Quick Start Invoke /flywheel-duel in an active flywheel session and let it detect the current phase, run the adversarial duel, and route the report into the matching pipeline.

Frequently Asked Questions about flywheel-duel

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

FAQPage Schema
How do I run a multi-agent adversarial duel in the flywheel workflow?

Invoke /flywheel-duel directly or reach it from a Discover, Plan, or Review menu. The skill pre-flights ntm and agent CLIs, auto-detects the duel mode from the current phase, runs the duel, and routes the report into the matching pipeline.

What agents are required to run a flywheel duel?

At least two of the claude, codex, and gemini CLI binaries must be available, plus the ntm orchestration tool passing its dependency check. If fewer than two agents are found, the duel is blocked and you can fall back to single-agent mode.

How does the duel mode get selected automatically?

The skill reads state.phase from .pi-flywheel/checkpoint.json and maps it to a mode: ideas for discovery phases, architecture for planning phases, and reliability for review phases. User-supplied --mode or --top flags always override the defaults.

What happens if the duel stalls or never finishes?

The skill polls the output report every 30 seconds up to a roughly 55-minute budget. On timeout or non-zero exit, it prompts you to salvage partial WIZARD_*.md transcripts, retry the duel, or abort and restore the pre-duel flywheel state.

Where do duel reports and transcripts get saved?

Reports are routed by phase into docs/discovery, docs/plans, docs/reviews, or docs/duels. The WIZARD_*.md transcripts stay in the project root as provenance for bead creation and are flagged stale by flywheel-cleanup after 7 days.