Full Automation Guide - LV CAD

🤖 Autonomous Development Pipeline

This repository now has fully autonomous development capabilities. The system can implement features with minimal human intervention.

🚀 Quick Start

Prerequisites Check

# Ensure these are running:
ollama serve                    # Start Ollama server
ollama pull deepseek-coder     # Ensure model is downloaded

Run Full Automation

# One-time run through all tasks
.\scripts\full_auto.ps1

# Continuous mode (daemon)
.\scripts\full_auto.ps1 -ContinuousMode

# Skip approval (auto-merge - not recommended)
.\scripts\full_auto.ps1 -ApprovalRequired:$false

📋 How It Works

1️⃣ Task Queue System

Tasks are stored in tasks/ directory:

task-*.md - High priority infrastructure tasks
feat-*.md - Feature implementation tasks

Example task file:

### Task: Create Database Connection Manager

**Objective:** Create centralized SQLite connection manager

**Key Steps:**
1. Create `db/connection.py`
2. Implement singleton pattern
3. Add tests

**Acceptance Criteria:**
- No startup errors
- Tests pass

2️⃣ Automation Pipeline

┌─────────────────────────────────────────────────────────┐
│                   FULL AUTO PIPELINE                    │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  1. 📋 Read Task Queue                                 │
│     └─ Get next unprocessed task from tasks/          │
│                                                         │
│  2. 🌿 Create Feature Branch                           │
│     └─ Branch: feat/auto-<task-name>                   │
│                                                         │
│  3. 🤖 AI Implementation                               │
│     ├─ DeepSeek Coder analyzes task                   │
│     ├─ Generates implementation plan                   │
│     ├─ Creates/modifies files                          │
│     └─ Generates pytest tests                          │
│                                                         │
│  4. 🧪 Automated Testing                               │
│     ├─ Black formatting                                │
│     ├─ Ruff linting                                    │
│     └─ pytest execution                                │
│                                                         │
│  5. 📤 Create Draft PR                                 │
│     └─ Labeled: automated, needs-testing               │
│                                                         │
│  6. ⏸️  HUMAN TESTING REQUIRED ⏸️                      │
│     ├─ Review code changes                             │
│     ├─ Manual testing in app                           │
│     └─ Approve PR on GitHub                            │
│                                                         │
│  7. 🔀 Auto-Merge                                      │
│     ├─ Wait for CI checks                              │
│     ├─ Squash merge to main                            │
│     └─ Archive task to tasks/pr/                       │
│                                                         │
│  8. 🔄 Repeat for Next Task                            │
│                                                         │
└─────────────────────────────────────────────────────────┘

3️⃣ Your Role (Human Testing)

When automation pauses for you:

Review the PR on GitHub
- Check code quality
- Verify logic correctness
- Look for edge cases

Test manually

# Checkout the PR branch
gh pr checkout <PR_NUMBER>

# Run the application
python app/main.py

# Test the new feature
# Verify no regressions

Approve on GitHub
- If tests pass → Approve PR
- If issues found → Request changes
- Automation will auto-merge after approval

🛠️ Advanced Usage

Single Task Implementation

# Implement one specific task
python scripts/ai_coder.py tasks/task-db-connection-manager.md

Add New Task

# Create task file
New-Item -Path "tasks/task-my-feature.md" -ItemType File

# Edit with your task description
# Follow format: Objective, Key Steps, Acceptance Criteria

# Automation will pick it up automatically

Monitor Progress

# Watch automation log
.\scripts\full_auto.ps1 | Tee-Object -FilePath logs/automation.log

# Check current PRs
gh pr list --label automated

# View task queue
Get-ChildItem tasks/*.md

🎯 Task Prioritization

Tasks are processed in this order:

task-*.md - Infrastructure/foundation tasks
feat-*.md - Feature implementation
Others - General improvements

Example Queue

tasks/
├── task-db-connection-manager.md          # ← Processed first
├── task-integrate-coverage-service.md     # ← Then this
├── feat-cad-core-trim-suite.md            # ← Then features
├── feat-backend-schema-loader.md
└── pr/                                     # ← Completed tasks archived here
    └── task-completed-example.md

🔧 Configuration

AI Model Settings

Edit scripts/ai_coder.py:

self.model = "deepseek-coder:latest"  # Change model
temperature = 0.3                      # Adjust creativity (0.0-1.0)
num_predict = 4096                     # Max tokens

Automation Behavior

Edit scripts/full_auto.ps1:

$CheckIntervalSeconds = 300      # How often to check for new tasks
$ApprovalRequired = $true        # Require human approval

📊 Monitoring & Metrics

View Automation Status

# Current branch
git branch

# Recent commits
git log --oneline -10

# PR status
gh pr status

# CI checks
gh run list --limit 5

Automation Metrics

The system tracks:

✅ Tasks completed
⏱️ Time per task
🧪 Test pass rate
📊 Lines of code generated

🚨 Troubleshooting

"Ollama not running"

# Start Ollama
ollama serve

# Verify
curl http://localhost:11434/api/tags

"DeepSeek Coder not found"

# Pull the model
ollama pull deepseek-coder

# Verify
ollama list

"Tests failing"

# Check the branch
git checkout feat/auto-<task-name>

# Run tests manually
pytest -v

# Fix issues, commit
git add -A
git commit -m "fix: address test failures"
git push

"AI generated bad code"

# Checkout the branch
git checkout feat/auto-<task-name>

# Edit files manually
# Fix the issues

# Commit and push
git add -A
git commit -m "fix: manual corrections to AI code"
git push

# PR will update automatically

🎓 Best Practices

Writing Good Task Files

### Task: [Clear, concise title]

**Objective:** One-sentence goal

**Key Steps:**
1. Specific, actionable step
2. Another specific step
3. Final step

**Acceptance Criteria:**
- Measurable outcome 1
- Measurable outcome 2

**Constraints:**
- Under 300 lines
- Add tests
- No breaking changes

Task Sizing

✅ Good: Single feature, <300 lines, clear scope
❌ Too Large: Multiple features, >500 lines, vague requirements
❌ Too Small: Trivial changes, not worth automation

Review Checklist

When reviewing AI-generated PRs:

🔐 Security & Privacy

Local AI Benefits

✅ Code never leaves your machine
✅ No cloud API costs
✅ Full control over models
✅ Works offline

GitHub Actions

Uses GitHub secrets for authentication
No code sent to external services
All CI runs in isolated containers

📈 Performance

Speed Estimates

Simple task (< 100 lines): 5-10 minutes
Medium task (100-300 lines): 15-30 minutes
Complex task (> 300 lines): Split into smaller tasks

Resource Usage

CPU: Moderate during AI generation
RAM: ~4-8GB for DeepSeek Coder
Disk: Minimal (models already downloaded)

🤝 Human-AI Collaboration

What AI Does Best

✅ Boilerplate code generation
✅ Test scaffolding
✅ Following established patterns
✅ Consistent formatting

What You Do Best

✅ Architecture decisions
✅ Edge case identification
✅ User experience evaluation
✅ Security review

The Sweet Spot

AI implements → You validate → AI refines → You approve

📞 Support

If Automation Gets Stuck

Check logs: logs/automation.log
Review GitHub Actions: gh run list
Check Ollama: ollama ps
Restart: Ctrl+C and re-run

Emergency Stop

# Stop automation
Ctrl + C

# Return to main
git checkout main

# Clean up branches
git branch -D feat/auto-*

🎉 Success Metrics

Track your automation success:

# PRs created by automation
gh pr list --label automated --state all

# Merge rate
gh pr list --label automated --state merged | wc -l

# Average time to merge
# (GitHub Insights > Pull Requests > Time to merge)

🔮 Future Enhancements

Planned improvements:

Multi-model support (GPT-4, Claude, etc.)
Self-healing (AI fixes its own test failures)
Intelligent task prioritization
Automatic documentation generation
Performance benchmarking

📚 Additional Resources

AI Model Docs: https://ollama.ai/library/deepseek-coder
Continue Extension: https://continue.dev
GitHub CLI: https://cli.github.com/manual/

Ready to start?

.\scripts\full_auto.ps1

Let the AI build while you focus on design and testing! 🚀

FilesExpand file tree

FULL_AUTOMATION_GUIDE.md

Latest commit

History