Creativity Lite

Research-backed creative thinking skill for Claude Code — fixation detection, constraint diversification, and incubation simulation in a standalone package.

What It Does

Creativity Lite is a structured creative analysis pipeline for Claude Code. It breaks AI fixation patterns and generates genuinely diverse hypotheses using techniques from 10 peer-reviewed papers (2025-2026).

Most LLMs follow the path of least resistance — they reliably produce good ideas, but those ideas cluster in the same conceptual space. Creativity Lite detects this clustering (fixation) and breaks it through constraint diversification and cognitive mode shifts.

Key Mechanisms

1. Constraint Diversification — Two parallel agents receive prompts built on different creative problem-solving paradigms (TRIZ contradiction resolution vs. Design-by-Analogy). Different constraints activate different cognitive search patterns, producing ideas from separate regions of the solution space.

2. Incubation Reset — Before idea generation, the system forces a problem restatement from a deliberately different angle. This simulates cognitive incubation — the phenomenon where stepping away from a problem leads to breakthroughs.

3. Fixation Detection — After generation, a diversity metric scores whether hypotheses actually cover distinct conceptual domains and approach types. If they cluster (score < 3/5), a defixation round injects ideas from an unrepresented domain.

Architecture

Phase 1  ACTIVATION         Extended thinking — parse themes, tensions, blind spots
   ↓
Phase 2  EXPLORATION        2-3 targeted web searches (trend, contrarian, adjacent)
   ↓
Phase 2.5 PROMPT OPT        Generate 2 prompts with DIFFERENT constraint paradigms
   ↓                        ├─ Prompt A: TRIZ-CONTRADICTION (systematic innovation)
   ↓                        └─ Prompt B: DESIGN-BY-ANALOGY (biomimicry-style mapping)
   ↓
Phase 3.0 INCUBATION        Force problem restatement → cognitive mode shift
   ↓
Phase 3  GENERATION         2 agents in parallel, each with its paradigm
   ↓                        ├─ @arch_sa (Sonnet) — TRIZ hypotheses
   ↓                        └─ @content_ts (Opus) — Analogy hypotheses
   ↓
Phase 3.1 FIXATION CHECK    Diversity score (1-5) → defixation if < 3
   ↓
Phase 3.5 SCORING           Novelty × Feasibility evaluation
   ↓
Phase 4  SYNTHESIS           Creative Brief with scored hypotheses + recommendation

Installation

# Copy the skill file to your Claude Code skills directory
cp SKILL.md ~/.claude/skills/creativity/SKILL.md

That's it. No dependencies, no MCP servers, no configuration.

Usage

In Claude Code, type:

creativity How might we redesign code review to catch architectural drift?

Options

• --distance=near — Analogies from the same industry (higher feasibility)
• --distance=medium — Analogies from different industries (balanced, default)
• --distance=far — Analogies from completely unrelated domains (higher novelty)

Example

creativity --distance=far What business models will emerge from AI agents?

Output

The skill produces a Creative Brief containing:

• Research Signals — Key findings from web research with sources
• Scored Hypotheses — 6-8 ideas evaluated on novelty and feasibility
• Diversity Assessment — Whether ideas genuinely cover different conceptual territory
• Provocative Questions — Challenges to conventional thinking
• Creative Recommendation — Best direction to pursue, backed by evidence

Research Foundation

This skill is grounded in 10 peer-reviewed papers from 2025-2026:

Paper	Key Finding	How We Use It
AutoTRIZ 2025	Automated TRIZ reasoning pipeline for LLMs	TRIZ contradiction prompts
TRIZ Agents 2025	Multi-agent LLM for TRIZ-based innovation	Multi-agent architecture
LLM-SCAMPER 2025	Idea quantity 9.4 vs conventional 2.7	Constraint diversification rationale
HAI-CDP 2025	Evolutionary generate-evaluate-mutate cycle	Fixation detection concept
AI Creativity Ceiling 2026	LLMs exceed average creativity but fixate	Core problem statement
Divergent-Convergent 2025	Different cognitive modes need different control	Incubation reset
Constraints as Catalysts 2025	Tight constraints → more innovative solutions	Constraint paradigm approach
Divergent Creativity in LLMs 2025	LLM divergent thinking characteristics	Diversity metric design
AI + Creative Process 2025	Human-AI creative collaboration patterns	Pipeline design
Generative System 3 2025	Systematic creativity frameworks	Overall methodology

Full Version

Creativity Lite is the open-source core of a larger system. The full Creativity v3 adds:

Feature	Lite	Full v3
Constraint paradigms	2 (TRIZ, Analogy)	3 (+SCAMPER-Absurdist)
Generation agents	2 parallel	3 parallel (+ wild card)
Fixation detection	Claude main evaluates	Dedicated agent evaluator
Stress testing	Simple scoring	Adversarial debate protocol
Evolutionary loop	—	COMBINE / INVERT / EXTREMIFY mutations
Knowledge Graph	—	Cross-session cumulative memory
Visual output	Table	Comparison grid + ASCII + Mermaid
Gemini cross-validation	—	Optional multi-model verification
Temporal lens	—	Past / present / future analysis
CLI flags	--distance	--distance, --evolve, --wild

Read the full article on how and why this was built: Engineering Creativity on Substack (link updated after publication)

Requirements

• Claude Code CLI
• No additional dependencies

License

MIT — see LICENSE

Author

Luigi Greco — AI systems architect, building tools for engineered creativity.

• Substack: luigigreco.substack.com
• GitHub: Luigigreco