v1.0 · MIT License // dlmastery · Claude Code Skill Suite

Meta‑Innovation
Harness

Four frontier LLMs. Six innovation workflows. One orchestrating mind. A novel blanking + leaking cross-pollination method that engineers Group Genius on any research goal.

4 Frontier AIs
6 Workflows
8 Skills total
∞ Research goals
Pitch Deck · 100 Slides

The Spectacular Deck

A 100-slide, auto-advancing explainer of the entire harness — the problem, the science of Group Genius, the blanking + leaking invention, the four experts, the six workflows, the meta / meta-meta layers, the guardrails, and the live proof. Plays automatically below; loops every ~6 seconds.

AUTOPLAY · LOOP 100 slides · 10 acts · dark / amber / cyan
The Four Polymaths

Fixed-Lens Expert Ensemble

Each frontier LLM receives a fixed disciplinary lens — a permanent epistemic stance it cannot abandon. The tension between lenses is the engine of novelty.

🧬
Claude
Anthropic · Biological Philosopher
  • Evolutionary Biology & Systems Biology
  • Ancient Texts & Classical Wisdom
  • Continental Philosophy & Ethics
  • Pattern languages of living systems
⚛️
ChatGPT (GPT)
OpenAI · Physical Engineer
  • Theoretical & Applied Physics
  • Chemistry & Material Science
  • Systems Engineering & Mechanism
  • Quantitative first-principles modeling
📊
Gemini
Google DeepMind · Human-System Evaluator
  • Social Science & Behavioral Economics
  • Product Strategy & Market Dynamics
  • HCI & User Experience Research
  • Evaluation methodology & measurement
🌀
Grok
xAI · Radical World-Modeler
  • Creative world-modeling & scenario design
  • Radical efficiency & constraint-stripping
  • Contrarian hypothesis generation
  • Science fiction as predictive methodology
Six Essential Questions

What · Why · Who · Where · When · How

01
What?
A Collective Intelligence Engine for AI
The Meta-Innovation Harness is a structured Claude Code skill-set — six "Innovation Harness" workflows, a meta-skill (auto-selects the right workflow), and a meta-meta-skill (generates & self-improves harnesses) — that directs four browser-driven frontier LLMs as a coordinated expert ensemble. The worked example targets six open research questions from a Google DeepMind/Gemini leadership talk.
02
Why?
Because No Single Mind Is Enough
Real breakthroughs emerge at disciplinary intersections — biology informing physics, philosophy correcting engineering. The Transactive Systems Model of Collective Intelligence (TSM-CI) and "Group Genius" theory show that structured cross-pollination among diverse experts outperforms any individual, including the best individual AI. This harness operationalises that theory as a repeatable, automatable protocol.
03
Who?
Orchestrator + Four Polymath Super-Experts
An Orchestrator AI (Claude Code, operating as a Triple-PhD MIT Genius) conducts the session. The four frontier LLMs — Claude, ChatGPT/GPT, Gemini, and Grok — each embody a permanent disciplinary stance. Intended users: researchers, strategists, innovation leads, and AI-augmented knowledge workers who need structured ensemble cognition.
04
Where?
Claude Code + Browser via Claude-in-Chrome
Runs entirely inside Claude Code CLI. All four frontier LLMs are driven through the Claude-in-Chrome browser extension — no APIs, no rate limits, no vendor lock-in beyond the four public chat interfaces. Outputs are archived as Markdown under examples/ in the repo for full transparency and reproducibility.
05
When?
Any High-Stakes Research Decision
Deploy when a question is too multi-dimensional for a single expert: frontier science, product strategy, policy, speculative engineering, or moonshot ideation. The inner/outer loop structure scales from a 30-minute "flash team" sprint to a quarterly deep-dive review lasting hours — chosen by the meta-skill based on goal complexity and time budget.
06
How?
Blanking + Leaking Cross-Pollination
For each focal question: blank (delete the prior answer for that question — forcing a clean-slate re-derivation) but leak (show the evolving best answers for all OTHER questions). Rotate across questions in inner loops, then synthesise in outer loops. This destroys anchoring bias while preserving cross-domain signal, producing genuinely novel combinatorial insight.
Operating Stance

The Persona

Triple-PhD MIT Genius
Elite of the Elite · Orchestrator Persona

Every session begins with a persona declaration: the Orchestrator operates as a Triple-PhD MIT Genius — elite of the elite, best on Earth in thought leadership, novelty, creativity, and extreme polymathy, drawing wisdom from ALL fields simultaneously. This is not mere motivational framing; it is a formal epistemic contract that shapes the depth of prompting, the breadth of reference, the courage to pursue genuinely radical ideas, and the ruthlessness of self-critique. The persona is inherited by each expert AI for the duration of their lens-specific contribution, then synthesised by the Orchestrator into the integrated output.

Six Workflows + Two Meta-Skills

The Innovation Harness Suite

Each workflow is a distinct orchestration protocol — different cadence, format, and cognitive pressure profile. The meta-skill auto-selects; the meta-meta-skill generates new ones.

// 01
Cross-Pollination Engine
Pure blanking + leaking across all four experts on a fixed question set. The foundational protocol: each expert erases their own prior and reads all others' evolving best answers before responding afresh.
Core Protocol
// 02
TSM-CI Kitchen Table
Simulates the Transactive Systems Model of Collective Intelligence as an intimate "kitchen table" dialogue — structured argument, role-acknowledgement, and shared memory building across the four expert nodes.
Theory-Grounded
// 03
AI-Delphi Swarm Hackathon
Multi-round structured expert elicitation (Delphi method) compressed into a hackathon cadence — rapid iteration, anonymous scoring rounds, and forced convergence toward testable hypotheses.
High-Velocity
// 04
Group Flow Improv Lab
Draws on Keith Sawyer's Group Flow research: improvisational constraints, "yes-and" extension rules, and real-time creative pressure to push all four experts beyond their safe answer envelopes simultaneously.
Creative Pressure
// 05
Ideaflow Flash-Team
Time-boxed sprint format (30–60 minutes) inspired by Jeremy Utley's Ideaflow methodology: maximum idea surface area in minimum time, with the Orchestrator as timekeeper and synthesiser.
Time-Boxed Sprint
// 06
Hybrid Quarterly Review
Long-horizon structured review: re-evaluates prior outputs from all five workflows against new evidence, expert critique, and domain updates. Closes the feedback loop between sprint and strategy.
Strategic Horizon
// Meta-Skill · Layer 7
Workflow Selector (Meta-Skill)
Receives a research goal + context → diagnoses the optimal workflow from the six above → instantiates and runs it end-to-end. Zero configuration required by the user; the meta-skill handles all parameterisation and expert briefing.
// Meta-Meta-Skill · Layer 8
Harness Generator (Meta-Meta-Skill)
Generates entirely new harness workflows, critiques them against collective-intelligence theory, and self-improves them. Turns the harness into a self-extending system — the skills write new skills. The ultimate expression of recursive innovation architecture.
System Architecture

The Blanking + Leaking Loop

Inner loops cycle across individual questions; outer loops synthesise across all experts and all questions. The Orchestrator AI holds the state machine.

// META-INNOVATION HARNESS · ARCHITECTURE v1.0 ORCHESTRATOR AI Triple-PhD MIT Genius Claude Code · Skill Runner CLAUDE-IN-CHROME Browser Automation Bridge CLAUDE Biology · Ancient Texts Philosophy Anthropic · Browser Tab ChatGPT / GPT Physics · Chemistry Engineering OpenAI · Browser Tab GEMINI Social Science · Product HCI · Evaluation Google DeepMind · Tab GROK World-Modeling · Radical Efficiency · Contrarian xAI · Browser Tab BLANK own answer · LEAK others' answers ARCHIVE → examples/ Full Markdown transparency OUTER LOOP · SYNTHESIS ↺ Inner loop: rotate across questions per expert · Outer loop: synthesise all
Theoretical Foundation

Why Blanking + Leaking Works

The method is grounded in cognitive science, collective intelligence theory, and information-theoretic principles of creative recombination.

Group Genius
Sawyer's Collaborative Emergence
Keith Sawyer's research shows that genuine creativity emerges from structured improvisation between diverse experts — not from individuals working alone. The harness encodes this as an algorithmic protocol.
TSM-CI
Transactive Systems Model
The Transactive Memory / Systems Model of Collective Intelligence holds that group IQ exceeds individual IQ when members hold complementary knowledge and communicate it efficiently. The four fixed lenses simulate this maximally.
De-Anchoring
Blanking Destroys Anchoring Bias
Cognitive anchoring — over-relying on the first answer — is the dominant failure mode of sequential LLM reasoning. Forcing deletion of the prior answer and mandatory re-derivation from scratch eliminates this bias completely.
Cross-Signal
Leaking Preserves Combinatorial Signal
While blanking eliminates anchoring within a question, leaking provides the cross-domain signal that drives genuine novelty. Seeing biology's answer before re-deriving the physics answer forces unexpected synthesis.
Delphi
Structured Iteration Toward Truth
Outer loops mirror the Delphi method's multi-round expert elicitation: each pass allows experts to update positions in light of synthesised group output, converging toward robustly defended consensus positions or surfacing genuinely irreducible tensions.
Worked Example
Six DeepMind Research Questions
The canonical demonstration applies the harness to six open frontier questions from a Google DeepMind/Gemini leadership talk — topics at the genuine edge of current AI research. Full archived outputs live in examples/.
Transparency & Lineage

Absolute Openness

Every interaction is archived. Every design choice is documented. Every intellectual ancestor is acknowledged.

📁
Full Session Archives
Every workflow run produces a complete Markdown transcript archived under examples/ — including all prompts, all expert responses, all Orchestrator synthesis steps, and all iteration traces. Science-grade reproducibility.
📐
Design Rationale
Every workflow design decision is explained in documentation/. The PRD, the theoretical grounding, the lens-assignment choices, and the meta-skill selection logic are all documented to PhD-thesis standard.
🔗
Intellectual Lineage
Built on the same author/persona's earlier work in automated research pipelines. See sibling repos below. Cites Sawyer, Woolley, Delphi method, TSM-CI, Ideaflow (Utley), and Group Flow literature throughout.
⚖️
MIT License
Fully open-source under the MIT License. Fork it, extend it, build on it. Contributions welcome — especially new workflow harnesses submitted via PR and validated against the meta-meta-skill's critique protocol.
Sibling Repositories · Same Author

The Broader Research Lineage

// Repository Structure
meta_innovation_harness/
├── skills/              # All 8 Claude Code skills (.md)
│   ├── 01_cross_pollination_engine.md
│   ├── 02_tsm_ci_kitchen_table.md
│   ├── 03_ai_delphi_swarm_hackathon.md
│   ├── 04_group_flow_improv_lab.md
│   ├── 05_ideaflow_flash_team.md
│   ├── 06_hybrid_quarterly_review.md
│   ├── 07_meta_skill_selector.md
│   └── 08_meta_meta_harness_generator.md
├── examples/            # Full archived session transcripts
├── research/            # Source questions, references, notes
├── documentation/       # PRD, theory docs, how-to guides
│   └── about/           # what / why / who / where / when / how
├── docs/                # GitHub Pages static site (this site)
└── assets/             # Shared assets