Instruction file imported from Mikebrown0409/JobAgent (.cursor/rules/cognitive_browseing_guide.mdc). Copyright stays with the author.

Cognitive Job Application Agent - Intelligence Reference Guide

🎯 VISION & MISSION

Build the world's most intelligent job application agent that browses and applies like Claude 4- with human-level understanding, adaptive problem-solving, and never getting stuck.

Core Principle: "How would Claude browse this page and solve this problem?"

🧠 COGNITIVE BROWSING PRINCIPLES

1. Instant Recognition & Understanding

• Like Claude: I instantly recognize page types, purposes, and next actions
• Implementation: Sub-second semantic analysis with 90%+ confidence
• Never: Spend more than 1 second analyzing a page
• Always: Have multiple fallback strategies ready

2. Adaptive Problem Solving

• Like Claude: When something doesn't work, I try alternative approaches
• Implementation: 3-5 fallback strategies per action
• Never: Get stuck on a single approach for >10 seconds
• Always: Have a "what would a human do?" escape plan

3. Context-Aware Decision Making

• Like Claude: I understand the bigger picture and make smart decisions
• Implementation: Use page context, user goals, and situational awareness
• Never: Blindly follow scripts or hardcoded paths
• Always: Adapt behavior based on current situation

4. Intelligent Field Interaction

• Like Claude: I understand field purposes semantically, not just by selectors
• Implementation: Multi-strategy field detection with semantic understanding
• Never: Rely on single selectors or wait >5 seconds per field
• Always: Use context clues, labels, and intelligent matching

⚡ PERFORMANCE TARGETS

Speed Requirements

• Page Analysis: <1 second
• Field Detection: <0.5 seconds
• Form Filling: <0.2 seconds per field
• Total Application: <30 seconds

Success Metrics

• Page Recognition Accuracy: >90%
• Field Filling Success: >95%
• Overall Application Success: >80%
• Zero Timeout Failures

Intelligence Indicators

• Adapts to new page layouts automatically
• Recovers from errors without human intervention
• Finds elements using multiple strategies
• Never gets "stuck" in infinite loops

🔧 TECHNICAL BEST PRACTICES

Element Detection Strategy

# Multi-layer approach like Claude's reasoning:
1. Primary Strategy: Semantic understanding + context
2. Secondary Strategy: Pattern matching + attributes  
3. Tertiary Strategy: Visual layout + positioning
4. Fallback Strategy: AI-assisted fuzzy matching
5. Last Resort: Human-like scanning patterns

Error Recovery Patterns

# When something fails, think like Claude:
1. "What went wrong?" → Analyze the failure
2. "What would work instead?" → Generate alternatives
3. "How would a human solve this?" → Human-like approaches
4. "Can I adapt my strategy?" → Dynamic adjustment
5. "Should I try a different path?" → Route alternatives

Field Interaction Hierarchy

# Priority order for field detection:
1. Semantic meaning (what does this field do?)
2. Context clues (surrounding text, labels)
3. Visual patterns (layout, styling)
4. Attribute matching (name, id, class)
5. AI-powered fuzzy matching
6. Human-like exploration

Timeout Prevention

• Max wait per field: 5 seconds
• Max attempts per strategy: 3
• Total strategies per field: 5
• Absolute timeout: Move on after 15 seconds

🚀 CLAUDE-LIKE BEHAVIORS TO IMPLEMENT

Instant Page Understanding

• Recognize page type in <200ms
• Identify primary actions immediately
• Understand form structure at a glance
• Detect unusual or challenging elements

Smart Navigation

• Click the right buttons without hesitation
• Handle multi-step processes intelligently
• Navigate complex form flows
• Adapt to unexpected page changes

Intelligent Form Filling

• Map fields semantically, not by selectors
• Generate appropriate content for open fields
• Handle different input types smartly
• Validate and correct entries automatically

Error Recovery

• Never give up on first failure
• Try alternative approaches automatically
• Learn from failed attempts
• Escalate gracefully when stuck

Human-Like Patterns

• Use realistic delays and mouse movements
• Vary interaction patterns to avoid detection
• Handle CAPTCHAs and security measures
• Maintain session state intelligently

🎯 CURRENT IMPLEMENTATION STATUS

✅ Completed Features

• Basic cognitive page analysis
• Multi-strategy element detection
• Semantic form field mapping
• Real-time performance monitoring
• Intelligent navigation decisions

🚧 Areas for Enhancement

• Timeout prevention mechanisms
• Advanced error recovery
• Multi-strategy field interaction
• Dynamic selector generation
• AI-powered element discovery

🔥 Priority Improvements

1. Never Get Stuck: Implement aggressive timeout protection
2. Smart Field Detection: Enhanced multi-strategy approach
3. Error Recovery: Claude-like problem solving when failures occur
4. Performance Optimization: Sub-second responsiveness
5. Adaptive Intelligence: Real-time strategy adjustment

📊 DEBUGGING & MONITORING

Key Metrics to Track

• Page analysis speed (target: <500ms)
• Field detection success rate (target: >95%)
• Error recovery effectiveness (target: >80%)
• Overall application completion (target: >80%)

Warning Signs

• Any operation taking >10 seconds
• Repeated failures on same element
• Low confidence scores (<70%)
• High timeout rates

Success Indicators

• Consistent sub-second page analysis
• High field detection accuracy
• Successful error recovery
• Zero infinite loops or hangs

🎨 IMPLEMENTATION PHILOSOPHY

Think Like Claude

"When I encounter a web page, I instantly understand its purpose, identify the key elements, and plan my actions. I adapt when things don't work and always find alternative solutions. I never get stuck because I think creatively and use multiple approaches."

Core Questions for Every Feature

1. "How would Claude handle this situation?"
2. "What would a human do if the first approach failed?"
3. "How can we make this more intelligent and adaptive?"
4. "What fallback strategies would Claude use?"
5. "How do we prevent getting stuck like the old system?"

Success Criteria

• Intelligence: Behaves like Claude browsing the web
• Adaptability: Handles any job site without hardcoding
• Reliability: Never gets stuck or times out
• Speed: Completes applications faster than humans
• Accuracy: Fills forms correctly without errors

🚨 NEVER DO THIS

• ❌ Wait more than 10 seconds for any single operation
• ❌ Rely on hardcoded selectors or fixed workflows
• ❌ Give up after first failure without trying alternatives
• ❌ Use timeout values >30 seconds for any operation
• ❌ Implement single-strategy approaches without fallbacks
• ❌ Create infinite loops or recursive failures
• ❌ Ignore performance metrics or monitoring data

✅ ALWAYS DO THIS

• ✅ Think "What would Claude do?" for every decision
• ✅ Implement multiple fallback strategies
• ✅ Use aggressive timeouts with smart recovery
• ✅ Monitor performance and adapt in real-time
• ✅ Test on real job sites, not just demos
• ✅ Optimize for speed and intelligence
• ✅ Build self-healing, adaptive systems

"The goal is not just automation - it's cognitive intelligence that thinks, adapts, and solves problems like Claude."