DSH_xieshujing

Managing a long-form novel is rarely about hitting a blank page; it is about pasting the whole book into a chat box round after round, juggling scattered files for outlines and chapter notes, and constantly worrying that an AI agent might silently overwrite a sentence you spent hours polishing. DSH_xieshujing (Write Whale) exists to kill those frictions: it is a fully DSH-native novel workbench where the human and the agent always edit the same book, without any copy-paste shuttle.\n\nThree pillars define what it does. First, WYSIWYG context: the chapter you are reading, the passage you have highlighted, and the reference you have pinned all become traceable creative context for the agent, while the chapter outline sits right beside the prose so you can unfold it when stuck and tuck it away when the flow returns. Second, a dedicated novel specialist: a full set of Novel Tools, Skills, Subagents, and an Agent Preset lets the agent work around outlines, chapter outlines, Story State, and versioned proposals rather than generically continuing text. Third, self-evolution: every manual edit, prompt-driven prose change, and final chapter lock is logged and analyzed so Write Whale gradually learns your taste in writing. After a chapter is done you can run a strict review or a NOAI scan that flags logic gaps, pacing issues, character inconsistencies, and mechanically repetitive phrasing bound to the current version, but both only diagnose and never rewrite behind your back. Every agent modification arrives as a diff you can accept, reject, revise, or roll back, and only versions you explicitly mark as final feed into the preference-learning loop.\n\nWrite Whale is built for authors serializing a long novel who want the agent to remember prior chapters and the book's own rhythm; for DSH writers who are done with chat logs plus a pile of loose files; for anyone who wants AI in the loop for continuation, rewriting, and proofreading while keeping every change visible, selectable, and restorable; and for writers who prefer their manuscript to stay on their machine, managed with Markdown, YAML, and Git. Your book stays yours.