What changed
LDF v2 is a complete repositioning. It is no longer a hook toolkit. It is a methodology for building well-architected codebases with AI assistance.
New core ideas
- Hexagonal architecture as the structural foundation
- Pure assemblies with compiler-enforced boundaries
- Invariant-driven development (INV-xxx codes, citable, verifiable)
- Framework over hardcoding (use established systems, not magic numbers)
- AI instruction files (CLAUDE.md — most impactful file in any repo)
- Ratchet testing (metrics that can only improve)
- Intentional knowledge hardening (update docs when it matters, not on every commit)
Roslyn: semantic code understanding
Major new section on using Roslyn for C#/.NET codebases. Grep finds strings. Roslyn sees the lifeblood flowing through the codebase. Every language needs this.
Two AI agents + human
- Code agent (embedded): full codebase access, maintains docs, runs verification
- Chat agent (disconnected): fresh codebase every conversation, spots what the code agent misses
- Human: architects, decides when to harden knowledge
Proven on DAWG
406,000 lines of C#. 51 assemblies, 13 pure. 76 invariants across 35 scopes. Solo dev, 5 months.
Repo restructured
- Short README (150 lines) as landing page
- Deep-dive pages: Principles, Roslyn, Workflow, Case Study
- Removed dead weight (empty hooks, unused dependencies, legacy examples)
- All docs aligned with new direction
How to use: Read the principles. Tell your AI agent to read them too. Implement what makes sense. The hooks are a bonus, not the point.