dueling-idea-wizards
Made by JeffMulti-agent adversarial idea generation and scoring via NTM swarm. Use when "dueling idea wizards", "competing ideas", "adversarial brainstorm", or cross-model idea evaluation for any project.
Works With
Quality and Utility come from canonical skill metadata. Community ratings appear separately as stars when subscriber reviews exist.
Subscribe to download
Public previews are available, but downloads, CLI sync, and premium content require an active subscription.
Sign In to Subscribe- Full skill content with deterministic versioning
- CLI sync + local search index
- Access to Jeffrey’s premium library
Save to your account
Saved skills live in your cloud list. Run jsm sync on each machine to pull them locally.
What's New with this Skill
3Updates, enhancements, and release notes for this skill.
Enhanced: dueling-idea-wizards, five waits get their duration units
Every phase gate in the duel blocks on ntm --robot-wait, and current NTM rejects bare integers for --timeout. All five waits now carry explicit units (120s, 180s, 300s), so the duel no longer dies between ideation, scoring, and reaction phases with a missing-unit error.
- Phase gates at spawn, study, ideation, scoring, and reaction all pass unit-suffixed timeouts that the live binary accepts
Enhanced: Dueling Idea Wizards
The skill now targets Antigravity (agy), Google's successor to the Gemini CLI, as its Gemini-family dueling agent, so the 3-way duel keeps working with current tooling.
- gmi (legacy Gemini CLI) → agy (Antigravity CLI) across the whole skill
- agy still runs Gemini models; all Gemini scoring/personality guidance still applies
- No more Flash-downgrade watch: ntm hard-pins --agy to Gemini 3.1 Pro (High)
- Updated install command, detection, spawn flags, and output-file naming (WIZARD_IDEAS_AGY.md)
- Note: ntm's --agy is on main, not yet in a tagged release (latest v1.18.3)
Google replaced the standalone Gemini CLI (gmi) with Antigravity (agy), and ntm now supports agy as a full agent type. dueling-idea-wizards has been updated end-to-end: detection, spawn flags, install hints, pane labels, output-file and score-matrix tokens, and the troubleshooting guidance all now reference agy. Because agy still runs Gemini models (ntm pins it to Gemini 3.1 Pro (High)), every Gemini scoring/personality note in the skill still applies, and there's no longer a Flash-downgrade to watch for. Heads-up: ntm's --agy support is currently on ntm's main branch and not yet in a tagged r
Dueling Idea Wizards
Pit two different AI models against each other in adversarial idea generation. Each model independently proposes its best improvements, then scores the opponent's ideas 0-1000, then sees how the opponent scored theirs. The catty disagreements are the point: where models strongly agree despite different biases, the ideas are genuinely good. Where they trash each other, the ideas are suspect.
- 11-phase adversarial pipeline: study, ideate, cross-score, reveal, rebuttal, steelman, blind-spot probe, synthesis, bead creation
- Automatic agent detection (cc, cod, gmi) with 2-way and 3-way duel support
- 6 domain modes beyond the default ideas duel: architecture, security, UX, performance, reliability, innovation
- 373-line prompt bank with 16 prompt types including devil's advocate and convergence refinement
- Score matrix with consensus/contested/killed verdicts and post-reveal adjustment tracking
- Full integration pipeline: codebase-archaeology to duel to beads to implementation swarm to release
- 228-line adversarial dynamics guide: model personality profiles, 5 disagreement types, reveal reaction taxonomy
Single-model brainstorming has a basic flaw: the same model that generates the idea also evaluates it. Dueling Idea Wizards addresses this by spawning a Claude Code + Codex (or Gemini) swarm via NTM and running Jeffrey Emanuel's adversarial cross-scoring methodology end-to-end.
The full pipeline runs 11 phases: project study, independent 30-to-5 winnowing, cross-scoring with 0-1000 granularity, the reveal (where each model sees how the other judged its ideas), optional rebuttal rounds, steelman challenges (forced to argue FOR the opponent's best idea), blind spot probes (what did neither of u