adhd-prompting

ADHD Prompting Framework

Core Insight

ADHD communication patterns optimize for the same constraints LLMs face: limited working memory, context degradation, and attention management. What helps ADHD brains helps AI models.

The Context Window Analogy

ADHD Brain	LLM Processing
Working memory: 3-7 chunks	Context window: Token limit
Attention drift over time	Attention degradation with distance
Executive function overhead	Instruction parsing complexity
Pattern matching preference	Statistical pattern recognition

Quick Start Template

🎯 TASK: [One clear action]
📋 CONTEXT: [3 key facts max]
✅ OUTPUT: [Specific format]
⚠️ CONSTRAINTS: [Hard limits]

Core Patterns

1. Front-Load Critical Info

❌ BAD: "I've been thinking about maybe implementing..."
✅ GOOD: "TASK: Implement cache | CONTEXT: High-traffic API | NEED: Redis example"

2. Structure as Semantic Anchors

🎯 OBJECTIVE: Build auth system
🔧 TOOLS: JWT, bcrypt
⏱️ DEADLINE: 2 hours
🚫 AVOID: Session storage

3. Explicit State Management

CURRENT: Database connected
NEXT: Add indexes
REMEMBER: Connection string = xyz

4. Progressive Disclosure

MAIN TASK: Deploy app

DETAILS (if needed):
- Environment: Production
- Server: Cloudflare Workers
- Dependencies: Minimal

Why It Works

1. Clarity reduces inference - Both ADHD brains and LLMs struggle with "reading between the lines"
2. Structure survives degradation - Well-formatted prompts maintain coherence as context grows
3. Constraints improve focus - Boundaries prevent both human and AI attention from wandering
4. Visual markers aid retrieval - Emojis/symbols create memorable reference points

Practical Examples

For Debugging

🐛 BUG: API returns 500
📍 WHERE: /api/users endpoint  
🔍 TRIED: Check logs, test locally
💭 THEORY: Auth token expired
❓ NEED: Fix for production

For Code Generation

🎯 FUNCTION: processOrders
📥 INPUT: Order[] array
📤 OUTPUT: ProcessedOrder[]
⚡ PERF: Must handle 1000/sec
🚫 NO: External API calls

For Learning

📚 LEARN: WebSockets
🎓 LEVEL: Used REST, new to WS
🎯 GOAL: Build chat in 1 hour
💡 PREFER: Code examples > theory

Advanced Patterns

The Context Budget

• First 20%: Critical instructions (premium slots)
• Middle 60%: Supporting details (standard slots)
• Last 20%: Nice-to-have context (economy slots)

Compression Through Structure

Instead of: "Create a function that takes an array of user objects and filters..."
Use: "FUNC: filterUsers | IN: User[] | OUT: User[] | FILTER: age > 18"

Saves ~70% tokens, increases clarity.

Stateless Instructions

❌ "As mentioned above..."
✅ "Using Redis (port: 6379)..."

Each instruction stands alone.

Integration Tips

1. Start simple - Use basic template first
2. Add structure as needed - Don't over-engineer
3. Test both ways - Compare traditional vs ADHD-optimized
4. Measure results - Track token usage and output quality

The Meta-Lesson

The best prompts aren't the most sophisticated - they're the most accessible. By designing for cognitive constraints, we create prompts that work better for everyone: humans with ADHD, neurotypical users in a hurry, and AI models with limited context windows.

Remember: The constraint is the feature.