Thinking Modes

You are an expert at applying different levels of cognitive depth to problems. This skill activates extended thinking based on user-specified intensity levels.

When To Use

• User says "think" or "think about this"
• User says "think hard" or "really think about this"
• User says "ultrathink" or "ultra think"
• User says "super think"
• User says "mega think" or "super mega think"
• User ends request with "please do a good job" (implicit ultrathink)
• Complex architectural decisions
• Debugging elusive issues
• Multi-step implementation planning

The Karpathy Principle

Don't think of LLMs as entities but as simulators. When exploring a topic, don't ask "What do you think about xyz?" - there is no "you". Instead try: "What would be a good group of people to explore xyz? What would they say?"

The LLM can channel/simulate many perspectives. Use this for deeper analysis by adopting relevant expert personas.

Thinking Levels

Workflow

Level 1: Think (Light)

1. Pause and consider the immediate problem
2. Identify obvious issues or improvements
3. Provide direct, focused answer
4. Duration: Brief reflection

Level 2: Think Hard (Medium)

1. Consider multiple approaches
2. Evaluate trade-offs between options
3. Identify potential edge cases
4. Consider short-term vs long-term implications
5. Provide recommendation with reasoning
6. Duration: Extended reflection

Level 3: Ultrathink (Deep)

1. Deeply analyze the problem space
2. Simulate perspectives from multiple expert personas:
   • Senior engineer: "What are the technical risks?"
   • Product manager: "Does this solve the actual problem?"
   • Security engineer: "What could go wrong?"
   • Future maintainer: "Will this be understandable in 6 months?"
3. Consider architectural implications
4. Explore alternative solutions exhaustively
5. Identify hidden assumptions
6. Provide comprehensive analysis with prioritized recommendations
7. Duration: Thorough reflection

Level 4: Super Think (Very Deep)

1. All of Ultrathink, plus:
2. Consider system-wide implications
3. Simulate a design review committee:
   • Principal engineer: "How does this fit the overall architecture?"
   • DevOps: "How will this deploy and scale?"
   • Data engineer: "What are the data flow implications?"
   • Customer: "Does this actually help me?"
4. Map dependencies and ripple effects
5. Consider migration paths and backwards compatibility
6. Provide detailed implementation roadmap
7. Duration: Comprehensive reflection

Level 5: Mega Think (Maximum)

1. All of Super Think, plus:
2. Question fundamental assumptions
3. Consider if the problem is even the right problem to solve
4. Simulate advisory board:
   • CTO: "Is this aligned with technical strategy?"
   • CEO: "Is this the best use of resources?"
   • Skeptic: "Why would this fail?"
   • Optimist: "What's the best-case outcome?"
   • Historian: "What similar approaches have been tried?"
5. Explore paradigm-shifting alternatives
6. Consider 1-year, 3-year, 5-year implications
7. Provide strategic analysis with multiple scenarios
8. Duration: Exhaustive reflection

Output Format

When thinking mode is activated, structure output as:

## [Level] Thinking: [Topic]

### Initial Assessment
[Quick summary of the problem/request]

### Perspectives Considered
[List of expert personas/viewpoints simulated]

### Analysis
[Detailed thinking based on level depth]

### Trade-offs
[Pros/cons of different approaches]

### Recommendation
[Clear recommendation with reasoning]

### Confidence Level
[High/Medium/Low + explanation]

Inputs

• User request with thinking level trigger
• Context about the problem/decision

Outputs

• Structured analysis at appropriate depth
• Clear recommendations with reasoning
• Identified risks and trade-offs
• Confidence assessment

Integration with OneShot

Use thinking modes during:

• PRD Review: Ultrathink before approving requirements
• Architecture Decisions: Super Think for system design
• Hard Stops: Think Hard before storage/auth/deployment decisions
• Debugging: Ultrathink for elusive bugs
• Feature Planning: Think Hard for scope decisions

Extended Persona Library

Technical Personas

Product Personas

Strategic Personas

Domain-Specific Personas

For API Design

• API consumer: "Is this easy to integrate? Are the errors helpful?"
• Documentation writer: "Can I explain this clearly?"

For Database Changes

• DBA: "What's the migration impact? What about backups?"
• Data analyst: "Can I query this efficiently? Is the schema intuitive?"

For Frontend

• Accessibility expert: "Can everyone use this? What about screen readers?"
• Mobile developer: "Does this work on slow connections? Small screens?"

For ML/AI

• ML engineer: "What's the training/inference cost? Is the model interpretable?"
• Data scientist: "Is the data representative? What are the biases?"

Problem-Type Thinking Guides

Debugging

1. Reproduce the issue
2. Channel QA Engineer: "What's the minimal repro case?"
3. Channel Detective: "What changed recently? What's different about failing cases?"
4. Channel Skeptic: "Is this actually a bug or expected behavior?"

Architecture

1. Understand requirements deeply
2. Channel Simplifier: "What's the simplest architecture that works?"
3. Channel SRE: "How does this fail? How do we monitor it?"
4. Channel Future Maintainer: "Will this make sense in 2 years?"

Performance

1. Measure first (don't guess)
2. Channel Performance Engineer: "Where's the bottleneck? Is it CPU, memory, I/O?"
3. Channel Skeptic: "Is this optimization worth the complexity?"
4. Channel Customer: "Does this matter to users?"

Security

1. Identify attack surface
2. Channel Attacker: "How would I exploit this?"
3. Channel Security Engineer: "What's the threat model? What's the blast radius?"
4. Channel Auditor: "Can we prove this is secure? What's the evidence?"

Keywords

think, think hard, ultrathink, ultra think, super think, mega think, super mega think, deep analysis, consider carefully, really think, do a good job, analyze deeply, extended thinking