### Added
- LLM fallback for unresolved conflicts (opt-in, experimental — multiple providers including Claude)
- Audit trail with prompt hash and validation score
- Settings panel for AI provider configuration
- MCP transport for self-hosted agents