Repository navigation
Releases: Xiaoyangy/novel-studio
Releases · Xiaoyangy/novel-studio
Release list
v0.3.0
Changelog
New Features
- 56a1b0c feat: add chapter-zero outline repair manifest
- 2084dea feat: add genre-aware prose and whole-chapter AIGC gates
- 3b70f21 feat: add independent character agent decisions
- 0b2206e feat: catch clinical/procedural register that reads as AI
- 55ce7ed feat: harden causal pipeline and golden-three rendering
- 9d637f9 feat: harden causal rewrite and exact-hash quality gates
- 4a60618 feat: harden pipeline and prose rendering
- 88db403 feat: harden render style pipeline
- 4a0ade0 feat: harden sealed short-story production pipeline
- 9e539f6 feat: model-allocated outline-all structure with no arc-length ceiling
- e63a8ea feat: persist pipeline planning and quality evidence
- 918db40 feat: plan and render novels one arc at a time
- f2a4029 feat: render plans as reader-facing scenes
- b0e75f5 feat: render prose reader-first with a readability signal
Bug Fixes
- fb8513c fix pipeline prose rendering and review gates
- a985313 fix: accept an identity-matched planning-context receipt without token echo
- fa2bd27 fix: accept assertion-negation reveal-budget in the project-all bundle gate
- ede7eb3 fix: accept assertion-negation reveal-budget items (不认定/不确认/…)
- 4003c7f fix: align grounded authority contract
- d04bc66 fix: align outline payoff evidence checks
- 9f129cb fix: bind plan world_simulation_id and protagonist_decision from the simulation
- 0c07fd9 fix: brighten chapter titles and align rag probes
- 0f38b01 fix: do not read "无论长篇还是短篇" as a longform declaration
- 7fea0ab fix: do not read "避免网络热梗" as a trend-language request
- ccd8361 fix: feed simulation errors into recovery
- 0144465 fix: harden project arc world simulation
- 2c56454 fix: harden short-story production pipeline
- a63953a fix: harden zero-init grounding contracts
- 6c3c9e6 fix: keep dashboard live during long pipeline calls
- 80005c8 fix: keep outline retries grounded and bounded
- 20fce3a fix: keep prose rendering out of workflow mode
- 6e85870 fix: make arc simulation recovery progress-aware
- 115c2d7 fix: make installation and startup reliable
- c2b90ff fix: parse whole-book chapter total in estimated_scale, not per-arc span
- 7bb55fb fix: preserve grounded authority through prompt compaction
- d2f55d6 fix: prioritize active simulation authority
- 4d3aa0b fix: raise the planning context budget to match the writer's window
- 4506c89 fix: replace stale dashboard instead of pinning to old code
- 8053f90 fix: resolve codex CLI bundled inside ChatGPT.app
- aed09ea fix: show per-chapter world-sim coverage, not just the arc-baseline tick
- 00c2135 fix: show planning and RAG dashboard progress
- d6fb772 fix: stabilize chapter rewrite and review pipeline
- 88c9d05 fix: stabilize local GGUF embedding batches
Others
v0.2.0
novel-studio v0.2.0 · Release 升级
这次 Release 升级把 novel-studio 从长篇生成 CLI 推进到一条可恢复、可验证、可观测的动态世界推演流水线。适合长篇网文、百万字连载、短篇投稿和自托管 AI 写作工作流。
Highlights
- Pipeline 正式升级为
architect -> zero-init -> write -> review -> rewrite -> deliver,阶段证据绑定 SHA-256,恢复和交付前都会复核产物新鲜度。 - 新增全角色章节世界推演:每章写作前先模拟角色目标、压力、资源、知识边界、误判、行动和蝴蝶效应,再投影到主角可感知正文。
- 返工链路绑定当前终稿 hash、审核 brief 和必须保留事实,避免 rewrite 阶段重造事件或正文长度失控。
- RAG/Qdrant 就绪链独立化,支持真实 embedding / vector store 状态检查、重试、恢复和轻量 index state 摘要。
- 审核链加入正文版本新鲜度、本地 AIGC 分片判定、DeepSeek 外部裁判、warning/blocking 分级和低风险非阻断策略。
- 进度看板 3.0 合并到主 README:新增总览、设定、人物、成长轨迹、计划、离屏世界、质量、运行八页签,并补充效果图。
- Release / update 文档前置:README 新增
Release 升级、GitHub Release 下载、一键安装、novel-studio update、Docker 和 Windows 升级入口。
Install / Upgrade
# macOS / Linux latest
curl -fsSL https://raw.githubusercontent.com/Xiaoyangy/novel-studio/main/scripts/install.sh | sh
# upgrade current binary to latest
novel-studio update
# install / upgrade to this release
curl -fsSL https://raw.githubusercontent.com/Xiaoyangy/novel-studio/main/scripts/install.sh | sh -s -- v0.2.0
novel-studio update v0.2.0Windows 用户请下载本页的 Windows ZIP,解压后替换 novel-studio.exe。
Docker:
docker pull ghcr.io/xiaoyangy/novel-studio:latestVerification
go test ./...goreleaser release --snapshot --clean --skip=publish