Skip to content

LLM-Assisted Development (LAD): Copilot + Claude Agent prompts & workflow for iterative Python feature implementation

License

Notifications You must be signed in to change notification settings

chrisfoulon/LAD

Repository files navigation

LAD — LLM-Assisted Development Prompt Kit

License: MIT

LAD enables systematic feature development and enterprise-grade test quality using Claude Code + GitHub Copilot Agent Mode. Build complex Python features iteratively and safely—from context gathering to 100% meaningful test success—with zero extra infrastructure.

✨ What's New in 2025

🔬 Enhanced Test Quality Framework — Achieve 90%+ test success through systematic PDCA cycles
🎯 Industry Standards Compliance — Research software + Enterprise + IEEE validation
📊 Session Continuity — Seamless interruption/resumption across multiple sessions
Real-World Insights — Based on 50+ LAD implementations in research software

Features

Test-driven development with atomic task breakdowns
Systematic test improvement with PDCA methodology
Component-aware testing (integration for APIs, unit for business logic)
Multi-level documentation with collapsible sections
NumPy-style docstrings enforced throughout
Session continuity with TodoWrite progress tracking
GitHub Flow with automated PR creation/cleanup
Agent autonomy with diff approval workflow

Choose Your Workflow

LAD supports two autonomous workflows optimized for different development environments:

🚀 Claude Code

Multi-phase autonomous workflow for command-line development

# Quick Setup
git clone --depth 1 https://github.com/chrisfoulon/LAD tmp \
  && rm -rf tmp/.git && mv tmp .lad \
  && git add .lad && git commit -m "feat: add LAD framework"

# Feature Development
git checkout -b feat/my-feature
# Tell Claude Code: "Use LAD framework to implement [feature description]"

Example: Starting a new feature

User: Use LAD framework to implement user authentication with JWT tokens

Claude: I'll use the LAD framework to implement user authentication. Let me start by reading the feature kickoff prompt.

[Claude automatically reads .lad/claude_prompts/00_feature_kickoff.md and begins setup]

🛠️ GitHub Copilot Agent Mode (VSCode)

Function-based autonomous workflow for IDE development

⚠️ Requires Copilot Agent Mode - standard Copilot Chat alone will not work with LAD

# Same LAD import as above
git checkout -b feat/my-feature
# Tell Copilot Agent: "Use LAD framework to implement [feature description]"

Example: Starting with Copilot Agent

User: Use LAD framework to implement user authentication with JWT tokens

Copilot Agent: I'll use the LAD framework for systematic implementation. Let me read the feature kickoff prompt and begin autonomous execution.

[Copilot Agent reads .lad/copilot_prompts/00_feature_kickoff.md and executes]

Framework Structure

.lad/
├── README.md                                     # This overview
├── LAD_RECIPE.md                                 # Complete workflow guide
├── claude_prompts/                               # 🚀 Claude Code workflow
│   ├── 00_feature_kickoff.md                     # Environment setup
│   ├── 01_autonomous_context_planning.md         # Context + planning
│   ├── 01b_plan_review_validation.md             # Optional validation
│   ├── 02_iterative_implementation.md            # TDD implementation
│   ├── 03_quality_finalization.md                # Final validation
│   ├── 04a_test_execution_infrastructure.md      # 🆕 Test execution setup
│   ├── 04b_test_analysis_framework.md            # 🆕 Pattern recognition
│   ├── 04c_test_improvement_cycles.md            # 🆕 PDCA methodology
│   └── 04d_test_session_management.md            # 🆕 Session continuity
├── copilot_prompts/                              # 🛠️ Copilot Agent workflow
│   ├── 00_feature_kickoff.md → 06_self_review_with_chatgpt.md
│   ├── 04a-04d_test_*.md                         # 🆕 Enhanced test quality
│   └── 04_test_quality_systematic.md             # 🆕 Single-file version
└── .vscode/                                      # Optional VSCode settings

Quick Examples

Feature Implementation (Phase 2 Continuation)

After planning is complete, continue implementation:

User: Continue with phase 2 implementation

Claude: I'll continue with the iterative implementation phase. Let me check the current TodoWrite status and proceed with the next pending task.

[Claude reads 02_iterative_implementation.md and resumes from current state]

Test Quality Improvement

Achieve systematic test improvement:

User: Use LAD test quality framework to achieve 100% meaningful test success

Claude: I'll use the enhanced test quality framework to systematically improve your test suite. Starting with phase 04a (Test Execution Infrastructure).

[Claude executes 04a→04b→04c→04d with PDCA cycles and user decision points]

Documentation

📖 LAD_RECIPE.md — Complete step-by-step workflow guide
🚀 Claude Code prompts — 7-phase autonomous workflow
🛠️ Copilot Agent prompts — Function-based autonomous workflow
🔬 Enhanced Test Quality — 4-phase systematic improvement framework

Requirements

For Claude Code Workflow

For Copilot Agent Workflow

  • VS Code with GitHub Copilot Agent Mode enabled
  • Python 3.11+
  • gh CLI for PR management (optional)

Code Quality Setup

LAD uses several tools to maintain code quality. Install them once per project:

pip install flake8 pytest coverage radon flake8-radon black

Both LAD workflows will guide you through creating .flake8 and .coveragerc configuration files during the kickoff process.

Workflow Characteristics

Both LAD workflows provide autonomous development with the same quality outcomes. Choose based on your development environment and preferences:

Claude Code Workflow

  • Environment: Command-line development with autonomous tool access
  • Interaction: Conversational with autonomous file operations
  • Context Management: Built-in tools for codebase exploration
  • Progress Tracking: TodoWrite integration with cross-session persistence

Copilot Agent Mode Workflow

  • Environment: VS Code IDE integration with agent capabilities
  • Interaction: Function-based development with structured prompts
  • Context Management: IDE file context with autonomous execution
  • Progress Tracking: Structured state management within development environment

Both workflows achieve the same outcomes — systematic feature development, comprehensive testing, and enterprise-grade quality — through different interaction models optimized for their respective environments.

Claude Code Workflow Phases

Core Development (Phases 0-3)

Phase Duration Capabilities
0. Feature Kickoff ~5-10 min Environment setup, quality standards, baseline metrics
1. Context & Planning ~10-15 min Autonomous exploration, TodoWrite breakdown, sub-plan evaluation
1b. Plan Review (Optional) ~5-10 min Cross-validation, quality assurance
2. Implementation (Resumable) ~30-120 min TDD loop, continuous testing, cross-session resumability
3. Finalization ~5-10 min Self-review, documentation, conventional commits

🆕 Enhanced Test Quality (Phases 4a-4d)

Phase Duration Capabilities
4a. Test Execution ~10-15 min Systematic chunking, timeout prevention, baseline establishment
4b. Test Analysis ~15-20 min Holistic pattern recognition, industry standards validation
4c. Improvement Cycles ~30-60 min PDCA cycles, TodoWrite integration, systematic fixes
4d. Session Management ~5-10 min Session continuity, context optimization, decision framework

Real-World Usage Patterns

Based on 50+ LAD implementations:

Session Management

  • Marathon Sessions (2-4 hours): Complex features with Phase 2 resumability
  • Focus Sessions (30-60 min): Test improvement cycles with PDCA methodology
  • Context Sessions (10-15 min): Session restoration and planning

TodoWrite Best Practices

  • Mark ONE task as in_progress before starting work
  • Complete tasks IMMEDIATELY after finishing
  • Break complex tasks into smaller, actionable items
  • Use descriptive task names for progress clarity

Test Quality Success Patterns

  • Start with P1-CRITICAL fixes (scientific validity + high impact/low effort)
  • Batch compatible fixes (infrastructure, API, test design changes)
  • Validate after each cycle (regression prevention essential)
  • User decision patterns: Most choose A (continue) after seeing progress

Context Optimization

Proven strategies for long sessions:

  • Use /compact <description> after major phase completions
  • Archive resolved issues before hitting context limits
  • Preserve successful patterns in CLAUDE.md
  • Session state files enable seamless resumption

License

This project is licensed under the MIT License.

Contributing

Improvements welcome! The LAD framework evolves based on real-world usage patterns and community feedback.

Framework Evolution Metrics:

  • Autonomous development workflows in both Claude Code and Copilot Agent Mode
  • 90%+ test success rates through systematic improvement methodology
  • Seamless session resumption across interruptions and context switches
  • Enterprise-grade quality standards with research software optimization

See LAD_RECIPE.md for complete framework details and contribution guidelines.

About

LLM-Assisted Development (LAD): Copilot + Claude Agent prompts & workflow for iterative Python feature implementation

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •