Ouroboros Formalizes Spec-First Agent Work With Numeric Gates — Ambiguity Score ≤0.2 Before Coding, Ontology Similarity ≥0.95 to Stop Looping

GitHub·medium signal

Q00/ouroboros (5,165 stars, +196 this week) replaces prompting with a five-phase loop: interview → seed → execute → evaluate → evolve. What distinguishes it from other spec-driven tooling is that the gates are numbers, not vibes — Socratic questioning must drive an ambiguity score to ≤0.2 before proceeding, and the evolve loop runs until ontology similarity converges at ≥0.95. The YAML seed spec is immutable once crystallized, execution uses Double Diamond decomposition, and evaluation is a three-stage gate of mechanical → semantic → multi-model consensus. Nine reasoning modes run across nine backends: Claude Code, Codex CLI, OpenCode, Hermes, Gemini, Kiro, Copilot CLI, Pi, and Zcode.