Integrate LAD (LLM-Assisted Development) Framework

[yarikoptic]

Not yet to merge but rather to consider . That's a framework I had mentioned earlier today

TODOs

• if to use -- needs to be tuned up: I will wait for upstream response on Overfits specific project kinds, and lacks instructions for customization chrisfoulon/LAD#4

Summary

This PR introduces the LAD framework to con-duct, providing systematic workflows for AI-assisted feature development using Claude Code and GitHub Copilot Agent Mode.

LAD enables test-driven development with structured phases, quality gates, and session continuity for sustainable development practices.

What is LAD?

LAD (LLM-Assisted Development) is a prompt-driven framework that provides repeatable workflows for implementing complex Python features iteratively and safely. It supports two autonomous development workflows:

• 🚀 Claude Code: 3-phase autonomous workflow optimized for command-line development
• 🛠️ GitHub Copilot Agent Mode: 8-step guided workflow for VSCode development

Key Benefits

✅ Systematic development with multi-phase autonomous workflows
✅ Test-driven development with atomic task breakdowns and continuous validation
✅ Quality assurance through enterprise-grade standards and automated gates
✅ Session continuity with TodoWrite progress tracking across interruptions
✅ Enhanced test quality via 4-phase PDCA (Plan-Do-Check-Act) methodology
✅ Component-aware testing strategies (integration for APIs, unit for business logic)
✅ Documentation standards with NumPy-style docstrings and multi-level docs

Framework Structure

The LAD framework is imported into .lad/ directory and includes:

.lad/
├── README.md                                   # Framework overview
├── LAD_RECIPE.md                               # Complete workflow guide
├── CLAUDE.md                                   # Persistent project context
├── claude_prompts/                             # 🚀 Claude Code workflow
│   ├── 00_feature_kickoff.md                   # Environment setup
│   ├── 01_autonomous_context_planning.md       # Context + planning
│   ├── 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_*.md         # 8-step guided process
│   └── 04_test_quality_systematic.md           # Enhanced test quality
└── .vscode/                                    # Optional VSCode settings

Usage Examples

With Claude Code

# After merging this PR
git checkout -b feat/my-feature

Then in Claude Code:

Use LAD framework to implement [feature description]

Claude automatically:

1. Reads .lad/claude_prompts/00_feature_kickoff.md
2. Sets up environment and quality baselines
3. Explores codebase and creates implementation plan
4. Implements feature with continuous testing
5. Validates and documents the implementation

With GitHub Copilot Agent Mode

Same setup, then in VSCode with Copilot Agent:

Use LAD framework to implement [feature description]

Copilot Agent executes the equivalent workflow using function-based prompts from .lad/copilot_prompts/.

Test Quality Improvement

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

Executes 4-phase systematic improvement:

• Phase 4a: Establishes comprehensive test baseline
• Phase 4b: Analyzes patterns across failure categories
• Phase 4c: PDCA cycles with user decision points
• Phase 4d: Session management and continuity

Real-World Validation

The LAD framework has been validated through:

• 50+ successful implementations across research software projects
• 90%+ test success rates through systematic improvement
• 3-5x faster development cycles via autonomous execution
• Seamless session resumption across interruptions and context switches

Integration Approach

This PR uses git subtree to import LAD framework:

git read-tree --prefix=.lad -u lad-import

The .lad/ directory is self-contained and version-controlled, allowing:

• Easy updates from upstream LAD repository
• Project-specific customizations in .lad/CLAUDE.md
• No external dependencies or build changes

Why LAD for con-duct?

con-duct is a research software project that benefits from:

1. Systematic quality standards: LAD enforces TDD, comprehensive testing, and documentation
2. Maintainability: Multi-level documentation and NumPy-style docstrings
3. Reproducibility: Session continuity enables picking up work across sessions
4. Test quality: Enhanced test framework ensures reliability of resource monitoring
5. Development velocity: Autonomous workflows speed up feature development

No Breaking Changes

This PR only adds the .lad/ directory. It does not modify:

• Source code in src/con_duct/
• Tests in tests/
• Build configuration (setup.cfg, pyproject.toml, etc.)
• CI/CD workflows
• Documentation in project root

The LAD framework is opt-in and used when contributors explicitly choose to follow LAD workflows for feature development.

Test Plan

• LAD framework files imported successfully
• No conflicts with existing project structure
• .lad/ directory self-contained
• Documentation files valid markdown
• No impact on existing tests: tox passes
• No impact on existing builds: pip install -e . succeeds

Documentation

The LAD framework is fully documented in:

• .lad/README.md - Quick overview and examples
• .lad/LAD_RECIPE.md - Complete step-by-step guide (550+ lines)
• .lad/CLAUDE.md - Project context and patterns
• Individual prompt files with detailed instructions

Future Work

After this integration, contributors can:

• Use LAD workflows for new feature development
• Adopt systematic test improvement for existing test suites
• Build up project-specific patterns in .lad/CLAUDE.md
• Customize workflows based on con-duct project needs

---

🤖 This PR enables systematic AI-assisted development for con-duct while maintaining full backward compatibility and zero impact on existing functionality.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.63%. Comparing base (3756590) to head (605e390).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #391   +/-   ##
=======================================
  Coverage   91.63%   91.63%           
=======================================
  Files          15       15           
  Lines        1112     1112           
  Branches      138      138           
=======================================
  Hits         1019     1019           
  Misses         70       70           
  Partials       23       23           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[yarikoptic]

seems to be overfit for some specific case already?