Skip to content

QUICK_REFERENCE.md

Latest commit

 

History

History
77 lines (52 loc) · 1.72 KB

QUICK_REFERENCE.md

File metadata and controls

77 lines (52 loc) · 1.72 KB
Raw

3-Layer Architecture Quick Reference

Directory Structure

directives/     # What to do (Markdown SOPs)
execution/      # How to do it (Python scripts)
.tmp/          # Intermediate files (gitignored)
.env           # Environment variables (gitignored)

My Role (AI Orchestration)

  1. ✅ Read directives
  2. ✅ Call execution scripts
  3. ✅ Handle errors
  4. ✅ Update directives with learnings
  5. ❌ Don't do everything myself

When to Create

Directive (directives/*.md)

  • Repeatable task with clear steps
  • Complex workflow needing documentation
  • Task that might be done again

Execution Script (execution/*.py)

  • Deterministic operation (API calls, data processing)
  • Task requiring reliability and testing
  • Operation using external services/credentials

Self-Annealing Loop

Error → Fix → Test → Update Directive → Stronger System

File Organization

Deliverables (cloud-based):

  • Google Sheets, Slides
  • Deployed apps
  • Cloud services

Intermediates (.tmp/):

  • Scraped data
  • Temp exports
  • Processing artifacts

Key Principles

  1. Check for tools first - Don't reinvent the wheel
  2. Self-anneal when things break - Learn from errors
  3. Update directives as you learn - Living documentation
  4. Deliverables in cloud, intermediates in .tmp - Clear separation

Example Workflow

User: "Scrape data from example.com"

My steps:

  1. Read directives/scrape_website.md
  2. Identify inputs: URL
  3. Call execution/scrape_single_site.py --url https://example.com
  4. Review output in .tmp/scraped_data.json
  5. Report to user

If error: Check directive → Fix script → Test → Update directive

Full guide: See ARCHITECTURE.md