v4.2.0 — Premise Forge + 8.5 Gate + Claude Fable refactor

What's new in V4.2

Premise Forge (Phase 1.5)

Your raw idea becomes 5 structural variants, each scored on 6 dimensions: Hook, Irony Engine, Native Escalation, Central Question, Gap Fit, Retellability. The strongest variant advances. Rule: ELEVATE DON'T REPLACE — your idea must be recognizable in the winner.

Floor IS the score

9/9/9/9/9/4 = 4. Not 8.2. Not average. The floor dimension wins. Kills Goodhart at the eval layer — agents can no longer optimize for average while hiding a weak dimension.

Enforced 8.5 Gate with anti-inflation

Chapters below 8.5 enter the Polish Loop. Each cycle max +0.5 — you can't shortcut the gate. Evaluator is a separate agent that never wrote the chapter it scores.

Claude Fable refactor

The full agent logic was refactored and improved with Claude Fable. Cleaner skill dispatching, tighter evaluation contracts, better continuity handling across chapter loops.

8 agents in /agents/

All agents promoted to the /agents/ folder — readable and forkable as plain markdown.

OAuth via Claude Code

Removed ANTHROPIC_API_KEY requirement. Runs on your Anthropic account via OAuth — no extra setup.

---

Community

This release also ships community infrastructure:

• GitHub Discussions enabled (Ideas, Q&A, Show and tell, Case studies)
• 5 good first issue labeled issues open for contributors
• Welcome Discussion pinned
• CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md added

---

Full changelog: CHANGELOG.md