← All use cases

中文

Research & writing · Claude Code

Draft a novel one chapter at a time with Claude Code slash commands

Idea + starter prompt, not run by us yet

For: Writers who want a first draft from an outline and have their own writing rules

The author had Claude Code turn itself into an assistant for writing a novel. It made about 10 slash commands, /brainstorm, /outline and /write-chapter among them. The author then pasted their own writing rules into the files. One /write-chapter prompt then produced 10 chapters in roughly 17 minutes, around 32,000 Chinese characters in total. The author called it a first draft.

Steps

  1. Ask Claude Code to set up brainstorm, outline and write-chapter commands and save them in .claude/commands.
  2. Paste your own writing rules into the command files before you run anything.
  3. Brainstorm ideas, pick one, then ask for a 10-chapter outline.
  4. Run write-chapter so each chapter reads the earlier text and the next outline entry, then saves as a txt file.
  5. Read it yourself. The author checked that a hint planted in chapter 2 paid off in chapter 9. Treat the whole thing as a first draft.

Starter prompt

Starter prompt. Change the [brackets] first.

Use only public material or a sanitized working copy. Use fictional client details. Do not read credential files or request secrets. If private client data appears, stop without quoting it.

Turn yourself into my assistant for writing a novel in this folder. Create the slash commands /brainstorm, /outline and /write-chapter and save them in .claude/commands. Show me the command files before you save them. My writing rules are: [paste your writing rules].

Genre and idea: [describe your idea]. First run /brainstorm and give me five ideas. Do not write any chapter until I pick one and approve the outline. Each chapter must read the earlier chapters and the next outline entry, and be saved as its own txt file in [chapters folder]. Never overwrite an existing file.

Original post: news.qq.com · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.