idea-darwin – Other skill for AI agents

Idea Darwin Engine

A round-based idea iteration system that treats ideas as competing organisms — scoring, selecting, crossing, and evolving them through structured rounds to surface the strongest concepts.

Overview

Most idea management tools are filing cabinets: they store ideas, tag them, and let them rot. Idea Darwin flips the paradigm — instead of organizing ideas, it lets them compete. Every idea is a living species on an evolution island. Each round, the fittest get deepened, different ideas cross-pollinate to produce unexpected hybrids, and external stimuli trigger mutations.

When to Use This Skill

• Use when you have many scattered ideas and need to systematically evaluate and develop them
• Use when you want to discover unexpected connections between ideas from different domains
• Use when you need structured iteration rather than one-shot brainstorming
• Use when you want a scoring framework to prioritize which ideas deserve more investment

Core Concepts

Evolution Island Metaphor

Your ideas are alive on this island. Like organisms, they follow three core laws:

1. Evolution — Each round, the system deepens the most viable ideas through structured research: filling logical gaps, clarifying paths, identifying risks.

2. Crossbreeding — The system cross-pollinates different ideas. A technical approach from work meets an observation from daily life, producing directions you never imagined.

3. Mutation — External stimuli (industry news, theories, conversations) trigger mutations, spawning entirely new species.

Species Cards

Every idea gets a structured card recording: core question, full description, lineage (parent/child IDs), 6-dimensional scores, and change history.

6-Dimensional Scoring

| Dimension | Weight | What It Measures |

|---|---|---|

| Novelty | 10% | Genuine breakthrough or repetition? |

| Feasibility | 20% | Technically and resource-wise achievable? |

| Value | 20% | Impact if successful? |

| Logic | 20% | Internally consistent, no gaps? |

| Cross Potential | 10% | Can spark something new when combined? |

| Verifiability | 20% | Can we design a validation path? |

Idea Lifecycle

seed → exploring → refining → crossing → validated → dormant

The user always has final say on all life-or-death decisions. The system only recommends.

Step-by-Step Guide

1. Write Your Ideas

Create an ideas.md file:

## Personal knowledge base that learns my style
I want a system that reads everything I write and gradually learns how I think.

## Commute-to-podcast converter
Record voice memos during my commute, auto-convert them into podcast scripts.

2. Initialize Your Island

The slash commands below are workflow request examples, not a CLI bundled with this catalog file. If the host has no command integration, perform the steps in Markdown: assign stable idea IDs, score each dimension on a stated 1–5 scale, compute the weighted mean, then propose selected refinements/crosses with lineage and the next validation experiment. Scores are judgments, not proof of validation. Mark an idea validated only after recording actual evidence.

/idea-darwin init

3. Start Evolving

/idea-darwin round

4. Keep Feeding the Island

Append new ideas to ideas.md, add environmental variables to stimuli.md.

Examples

Example 1: Initialize

/idea-darwin init --budget 8 --actions 3

Example 2: Run Multiple Rounds

/idea-darwin round 3

Example 3: Manage Ideas

/idea-darwin dormant IDEA-0005
/idea-darwin wake IDEA-0005

Best Practices

• Do: Write ideas as rough as you want — the system structures them
• Do: Add external stimuli to prevent idea convergence
• Do: Run disruption rounds to surface overlooked ideas
• Don't: Over-curate initial ideas — let evolution filter
• Don't: Ignore the "Decisions Needed" section in briefings

Additional Resources

• [GitHub Repository](https://github.com/warmskull/idea-darwin)
• Available in 3 languages: English, Chinese, Japanese
• ClawHub: clawhub install idea-darwin

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.