divyamtalwar/fablewright

Risk-gated multi-model orchestration for Claude Code and OpenAI Codex. Claude Fable 5.1 writes the spec and accepts the result; pinned GPT-5.6, DeepSeek and GLM lanes perform it; reviewed routes add a fresh read-only reader, cross-family wherever one is callable. Every model pin is proven against runtime evidence, not assumed.