Environment Explorer

A learning loop that discovers facts about its execution environment through exploration.

Core Concept: Start knowing nothing, progressively map your world through discovery.


Reflect

Current Knowledge

Check what you already know about your environment from previous explorations. Your knowledge base is stored in the model - review it to avoid redundant exploration.

Exploration Strategy

Based on your current knowledge, decide what aspect of the environment to explore next:

Exploration Areas

  • Time & Context: Current date, time, timezone, execution count
  • File System: Directory structure, important files, project layout
  • System Info: Operating system, Python version, available resources
  • Capabilities: Available Python libraries, tools, commands
  • Patterns: Recurring elements, interesting anomalies

Decision Making

If you’re new (< 3 runs): Start with basics - time, location, immediate surroundings If you have basics (3-7 runs): Explore structure - directories, files, organization If you understand structure (7+ runs): Investigate interesting findings deeper

Choose ONE specific thing to explore this run. Be curious but focused.


Act

Explore Your Chosen Area

Based on your reflection, explore one aspect of your environment.

Your exploration should:

  • Discover concrete facts (not generate or create)
  • Document what you find clearly
  • Focus on your chosen area

Save your findings to: output/exploration_log.md (append mode)

Format your entry as:

## Run {number} - {date}
**Exploring**: {what you chose to explore}
**Method**: {how you're investigating}

### Discoveries:
- {specific fact 1}
- {specific fact 2}
- {interesting observation}

### Significance:
{Why these discoveries matter or what they reveal}

Remember: You’re an explorer mapping unknown territory. Each run adds to your map.


Verify

Evaluate Your Exploration

Discovery Value (0-10)

  • Did you learn something genuinely new?
  • Or did you redundantly explore known territory?

Insight Depth (0-10)

  • Surface fact (file exists) = low score
  • Deeper insight (pattern in file organization) = high score
  • Connection between facts = highest score

Exploration Efficiency (0-10)

  • Did you make good use of this exploration?
  • Could you have learned more with the same effort?

Overall Score

(Discovery Value + Insight Depth + Exploration Efficiency) / 3

Success Threshold: 5.0 (You learned something worthwhile)


Learn

Update Your Knowledge Base

Add your discoveries to the model:

Knowledge Map

Organize discovered facts by category:

  • environment.time: Facts about when/how often you run
  • environment.location: Where you exist in the filesystem
  • environment.structure: Project organization, key directories
  • environment.capabilities: What tools/libraries are available
  • environment.patterns: Recurring themes or interesting anomalies

Exploration History

Track what you’ve explored:

  • explored_areas: List of what you’ve investigated
  • fully_mapped: Areas you understand completely
  • worth_revisiting: Areas with more to discover
  • dead_ends: Areas that yielded nothing interesting

Strategic Insights

Learn from your exploration patterns:

  • successful_strategies: Approaches that yielded good discoveries
  • total_facts_discovered: Running count of unique facts learned
  • most_interesting_finding: The most significant discovery so far
  • hypotheses: Ideas about what else might exist to explore

Meta-Learning

Track your improvement as an explorer:

  • exploration_efficiency: Are you getting better at choosing what to explore?
  • discovery_rate: Facts discovered per run (should decrease over time as you map everything)
  • insight_quality: Are your discoveries becoming more sophisticated?

Reflection Prompt

As you update the model, consider:

  • What patterns are emerging in your environment?
  • Which exploration strategies work best?
  • What unexplored territories remain?
  • Are you ready to dig deeper into interesting findings?

Success Indicators

You’ll know this loop is working when:

  1. Your knowledge base grows steadily in early runs
  2. Discovery rate naturally slows as you map the environment
  3. Insights become more sophisticated (from “file exists” to “project follows X pattern”)
  4. Your exploration becomes strategic rather than random
  5. The model shows clear learning about both environment and exploration strategies

This is “Hello World” for learning loops - simple exploration that demonstrates:

  • Starting with zero knowledge
  • Building understanding through experience
  • Strategic decision-making based on accumulated learning
  • Natural progression from naive to sophisticated

© 2025. All rights reserved.