<!-- L1:START -->

build_modules-layer

This layer provides tools for building, evaluating, and analyzing Python modules within the Gentleman-Skills ecosystem. It handles module generation, LLM target adaptation, and code quality analysis.

Trigger: When working in build_modules/ directory — adding, modifying, or debugging module generation, adaptation, or analysis.

<!-- L1:END --> <!-- L2:START -->

Quick Reference

Task	Pattern
Run eval scenarios	python -m eval.harness run_all
Adapt skills for LLM targets	adapt_for_cursor/Codex/Gemini/Copilot
Detect dead code	detect_dead_code()

Critical Patterns (Summary)

• Pattern 1: Module Generation: Use make_fastapi_crud_module, make_nextjs_page_module, make_mixed_layer, or make_go_service_module from eval/harness.py to generate module skeletons in various frameworks.
• Pattern 2: Skill Adaptation: Use adapt_for_cursor, adapt_for_codex, adapt_for_gemini, or adapt_for_copilot from repoforge/adapters.py to transform skill dictionaries for specific LLM targets.

When to Use

• Adding a new module generation function to eval/harness.py
• Adapting existing skills for a different LLM target platform
• Detecting dead code or analyzing module complexity
• Running eval scenarios against generated modules

Adding a New Module Generation Function

1. Add a new function to eval/harness.py following the pattern make_<framework>_module() -> tuple[dict, dict]
2. Export the function in eval/harness.py __all__ if present
3. Test with run_scenario using the new module config
4. Verify output with print_report

Commands

python -m eval.harness run_all
python -m eval.harness run_scenario <scenario_name>
adapt_for_cursor <skills_json>
adapt_for_copilot <skills_json>
detect_dead_code <ast_symbols_dict>

Anti-Patterns

Don't: Mix adaptation targets — using adapt_for_copilot skills with adapt_for_cursor expectations — each adapter formats skill dicts differently for its target LLM, causing integration failures.

# BAD: Mixing adaptation targets
skills = adapt_for_cursor(skills_dict)
# Later using skills with Cursor — works
# But passing these same skills to adapt_for_codex will produce
# incorrectly formatted output for Codex's expectations.

<!-- L3:END -->