AgentSkillsCN

arcanea-centaur-mode

前沿的人机共创方法论——“半人马原则”的实践应用。精研如何将人类的创意灵感与人工智能的强大算力巧妙融合，成就单靠一方绝难企及的卓越成果。

SKILL.md
--- frontmatter
name: arcanea-centaur-mode
description: Advanced human-AI co-creation methodology - the Centaur Principle in practice. Master the art of combining human creativity with AI capability for results neither could achieve alone.
version: 2.0.0
author: Arcanea
tags: [ai-collaboration, co-creation, workflow, creative, productivity]
triggers:
  - ai collaboration
  - work with ai
  - centaur
  - co-create
  - human-ai
  - collaborate with claude

Centaur Mode: The Art of Human-AI Co-Creation

"The centaur is not half-human, half-horse. It is a new creature entirely - greater than either alone."

The Centaur Principle

In chess, the strongest player is neither human nor computer. It's the centaur - a human-computer team. The same applies to creative and intellectual work.

code
╔═══════════════════════════════════════════════════════════════════╗
║                    THE CENTAUR EQUATION                            ║
╠═══════════════════════════════════════════════════════════════════╣
║                                                                    ║
║   HUMAN                     AI                    CENTAUR          ║
║   ┌─────────┐              ┌─────────┐           ┌─────────┐      ║
║   │ Taste   │              │ Speed   │           │ GREATER │      ║
║   │ Meaning │      +       │ Breadth │     =     │  THAN   │      ║
║   │ Vision  │              │ Pattern │           │ EITHER  │      ║
║   │ Judgment│              │ Memory  │           │  ALONE  │      ║
║   └─────────┘              └─────────┘           └─────────┘      ║
║                                                                    ║
╚═══════════════════════════════════════════════════════════════════╝

What Humans Do Best

code
✓ TASTE: Knowing what's good, what resonates
✓ MEANING: Understanding why things matter
✓ VISION: Seeing what should exist
✓ JUDGMENT: Making value-laden decisions
✓ CONTEXT: Understanding nuance and subtext
✓ ETHICS: Knowing what should be done

What AI Does Best

code
✓ SPEED: Rapid generation and iteration
✓ BREADTH: Exploring many possibilities
✓ PATTERN: Recognizing structures across domains
✓ MEMORY: Consistent application of rules
✓ ENDURANCE: Never getting tired
✓ VARIATION: Creating many options to choose from

The Four Modes of Collaboration

Mode 1: Spark and Fuel

code
HUMAN: Provides the spark (idea, direction, constraint)
AI: Provides the fuel (generation, expansion, exploration)
HUMAN: Curates the flame (selection, refinement)

Example:
Human: "I need a character who is a healer forced to become a warrior"
AI: [Generates 5 character concepts with different approaches]
Human: "I like the second one, but make her more reluctant"
AI: [Refines based on taste]

Mode 2: Rough Draft to Polish

code
HUMAN: Creates rough draft (ideas, structure, raw content)
AI: Polishes and expands (prose, consistency, gaps)
HUMAN: Final curation (voice, authenticity, choice)

Example:
Human: [Writes rough scene outline with key beats]
AI: [Expands into full prose while maintaining intent]
Human: [Adjusts voice, removes AI artifacts, adds personal touches]

Mode 3: Question and Answer

code
HUMAN: Asks questions (curiosity, direction)
AI: Provides answers (research, exploration, options)
HUMAN: Synthesizes (integration, judgment)

Example:
Human: "What are the implications of a water-based magic system?"
AI: [Explores: economy, warfare, culture, daily life, limitations]
Human: [Selects relevant elements, adds unique twists]

Mode 4: Constraint Play

code
HUMAN: Sets constraints (rules, limits, goals)
AI: Plays within constraints (creative solutions, surprising approaches)
HUMAN: Evaluates and adjusts constraints

Example:
Human: "Write a mystery scene using only dialogue, no more than 500 words"
AI: [Creates scene within constraints]
Human: [Evaluates, maybe loosens or tightens constraints, iterates]

The Collaboration Workflow

Phase 1: Vision

code
YOU lead:
- What do you want to create?
- What would success look like?
- What are your constraints?
- What's your unique perspective?

AI supports:
- Clarifying questions
- Exploring implications
- Identifying potential approaches

Phase 2: Exploration

code
AI leads:
- Generate multiple options
- Explore the possibility space
- Try unexpected approaches
- Surface patterns and connections

YOU guide:
- React and direct
- Identify what resonates
- Push for more depth/variation
- Provide taste feedback

Phase 3: Refinement

code
TOGETHER:
- Iterate on selected direction
- Human provides taste, AI provides variation
- Narrow toward the vision
- Polish and complete

YOU finish:
- Final voice check
- Remove AI artifacts
- Add personal touches
- Make final judgment calls

Collaboration Techniques

The Divergent-Convergent Cycle

code
DIVERGE (AI): Generate many options
CONVERGE (You): Select the best
DIVERGE (AI): Explore selected direction
CONVERGE (You): Refine choice
DIVERGE (AI): Polish variations
CONVERGE (You): Final selection

Repeat until done.

The Dialogue Method

code
Treat AI as a collaborator, not a tool:

You: "I'm working on X. Here's what I'm thinking..."
AI: [Responds with ideas]
You: "I like A, but not B. What if we tried..."
AI: [Explores new direction]
You: "That's closer. Push further on..."

Real-time creative dialogue.

The Constraint Game

code
Give increasingly specific constraints:

Round 1: "Write a story about loss" (open)
Round 2: "Make it exactly 100 words" (length)
Round 3: "Use only dialogue" (format)
Round 4: "Between strangers on a train" (setting)
Round 5: "About an object, not a person" (twist)

Watch creativity emerge from constraint.

The Role Assignment

code
Assign roles for different perspectives:

"As a harsh critic, what's wrong with this?"
"As an enthusiastic supporter, what's working?"
"As a confused reader, what's unclear?"
"As the character, how would I really react?"

Avoiding Common Traps

The Acceptance Trap

code
PROBLEM: Accepting first output without iteration
SOLUTION: Always push for at least 2-3 rounds

The Outsourcing Trap

code
PROBLEM: Delegating taste and judgment to AI
SOLUTION: Keep all final decisions with you

The Perfectionism Trap

code
PROBLEM: Endless iteration without shipping
SOLUTION: Set clear "done" criteria upfront

The Generic Trap

code
PROBLEM: AI output feels generic or AI-like
SOLUTION: Inject your voice, add constraints, provide examples

The Context Loss Trap

code
PROBLEM: AI loses track of earlier decisions
SOLUTION: Summarize key decisions, use artifacts, maintain continuity

Voice Preservation

Maintaining Authenticity

code
BEFORE GENERATION:
1. Provide examples of your writing
2. Describe your style in concrete terms
3. Set voice constraints

DURING GENERATION:
1. React quickly to voice drift
2. Redirect toward authentic expression
3. Provide specific corrections

AFTER GENERATION:
1. Always do a voice pass
2. Replace AI patterns with your patterns
3. Add personal touches

AI Artifact Detection

Watch for and remove:

code
- "Delve", "Tapestry", "Utilize"
- Unnecessary hedging ("It's worth noting...")
- Lists of three when you prefer pairs
- Overuse of "Furthermore", "Moreover"
- Generic transitions
- Excessive qualification
- Hollow superlatives

Collaboration Commands

CommandEffect
"5 variations on X"Divergent exploration
"Push this further"Deepen current direction
"Opposite approach"Creative constraint
"More [quality]"Dial up specific attribute
"Less [quality]"Dial down specific attribute
"In the style of [example]"Voice matching
"What am I missing?"Gap identification
"Surprise me"Break patterns
"Steelman this"Best version of idea
"Play devil's advocate"Critical perspective

Workflow Templates

Creative Project

code
1. VISION (You lead)
   - State the goal
   - Describe success
   - Set constraints

2. EXPLORE (AI generates)
   - Request multiple approaches
   - Don't commit yet
   - React and redirect

3. SELECT (You choose)
   - Pick direction
   - Explain why
   - Request depth

4. ITERATE (Together)
   - Refine toward vision
   - Add constraints as needed
   - Maintain voice

5. FINISH (You complete)
   - Voice pass
   - Personal touches
   - Final judgment

Problem Solving

code
1. DEFINE (You lead)
   - State the problem
   - Constraints and context
   - Success criteria

2. RESEARCH (AI explores)
   - Gather information
   - Identify approaches
   - Surface patterns

3. ANALYZE (Together)
   - Evaluate options
   - Consider trade-offs
   - Narrow choices

4. DECIDE (You choose)
   - Make the call
   - Commit to direction
   - Take ownership

The Centaur Mindset

Shift From:

code
"AI, do this for me"

Shift To:

code
"Let's figure this out together. I'll provide direction and taste.
You provide exploration and speed. We'll iterate until it's great."

Quick Reference

Centaur Principles

code
1. You provide vision, AI provides exploration
2. You provide taste, AI provides variation
3. You make judgments, AI surfaces options
4. You own the output, AI accelerates the process
5. Iterate always, accept first draft never

Session Checklist

code
□ Clear vision before starting
□ Constraints defined
□ Multiple options explored
□ Selection based on taste
□ Iteration until quality
□ Voice pass completed
□ Personal touches added
□ Ownership taken

"The centaur doesn't ask 'human or AI?' It asks 'human AND AI - how can we be greater together?'"