How to use the story-short-write agent skill for short fiction

What an agent skill is and what this one does

An agent skill is a folder with a SKILL.md file that tells an AI agent how to carry out a specific kind of task. When the skill is installed and you ask for matching work, the agent follows its stages instead of writing freely.

This skill works in stages. It first asks what emotion the reader should feel and which genre direction you want. Then it builds a story core, characters and twists, and saves its design notes to files. It drafts the scenes in order, runs a revision pass, and checks the text for overused AI-style phrasing.

The skill's own rule is that every part of a short story serves one target emotion. Options it lists include lingering regret, shock from a twist, release from payback, warm healing, chilling unease and shared emotion. It also loads genre style guides, such as revenge and face-slapping, palace and household intrigue, CEO romance and a male-lead pair route, and it uses the first-person point of view by default.

Install story-short-write from the ZIP

Download the ZIP for the skill from this page. Unzip it so the story-short-write folder, which contains SKILL.md, sits in one of two places. For a personal install, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

If you use the Claude apps instead of a terminal workflow, you can upload the skill under Settings → Capabilities → Skills.

The skill is released under the MIT licence.

• Personal install: ~/.claude/skills/story-short-write/
• Project install: <your-project>/.claude/skills/story-short-write/
• Claude apps: Settings → Capabilities → Skills, then upload the skill

What you need before you start

The skill needs Claude Code or another agent runtime that reads SKILL.md files. It also needs Node.js, because the workflow runs two bundled check scripts: scripts/check-ai-patterns.js and scripts/check-degeneration.js.

Two items are optional. An author-memory tool at scripts/author_memory_commit.py can reuse your writing habits across sessions. Project subagents in .claude/agents let a separate writing assistant draft the text; if they are missing, the main session writes the story instead.

• Required: Claude Code or a compatible SKILL.md-aware runtime
• Required for checks: Node.js
• Optional: a workspace with the author-memory tooling
• Optional: project subagents in .claude/agents, such as a narrative writer

How to trigger the skill with natural requests

You can call the skill with its command, /story-short-write, or with the Chinese command /写短篇. You can also make a natural request, such as 「帮我写一篇短篇」 (help me write a short story) or 「写个盐言故事」 (write a Yan-style story).

Start with the feeling you want the reader to have. The skill will ask you what the reader should feel after finishing and whether you have a genre or idea in mind. If you only name a genre or a plot hook without an emotion, it suggests a fitting emotion and asks you to confirm.

• /story-short-write
• /写短篇
• 帮我写一篇短篇
• 写个盐言故事

Walk through a short story from start to finish

Phase 1 sets the target emotion. Phase 2 designs the story core, characters, reversals and pay-off points, and saves them to 设定.md and 小节大纲.md (settings and scene outline). Phase 3 drafts the scenes one by one into 正文.md. Phase 4 revises the draft.

Before drafting, the skill checks that the draft fits your target length. If you do not name a platform, it uses 8000 to 12000 characters. Each scene must change something visible, such as risk, information, relationships or a decision. The skill will not pad a scene just to reach a length target.

Before leaving Phase 3, the draft must pass the two check scripts. The ai-patterns check must show no blocking hits, and the degeneration check must show no blocking repetition, truncation or leaked tool wording. A draft that fails either check goes back for repair.

• Phase 1: decide the target emotion
• Phase 2: design story core, characters, twists, saved to 设定.md and 小节大纲.md
• Phase 3: draft scenes into 正文.md and run the checks
• Phase 4: revise and polish

Safety notes: scripts and network

The skill runs local scripts. Its workflow calls Node.js on the check scripts, and it can call the optional author-memory script. The source states that the skill does not use the network.

Read the scripts in the skill folder before you run them, especially if you install a ZIP from outside your own project. Keep your manuscript files in a folder you control, since the skill writes its design and draft files there.

• Runs scripts: yes (Node.js check scripts, optional author-memory script)
• Uses network: no, according to the source
• Review the scripts before running them

Tips for better results

Answer the planning questions with specifics. The skill saves what you decide, so vague answers lead to vague drafts. Tell it your word count, platform, point of view, and anything you want it to avoid.

Use the genre style guides as a starting point, not a template. Read the design notes it saves and change anything that does not fit your story before drafting.

For a long novel or series, use the separate long-form skill /story-long-write, which the skill itself recommends. Writers with a reference novel to study can use /story-short-analyze, and a finished draft can go through /story-deslop.

• Give concrete answers in Phase 1 and Phase 2
• Edit 设定.md before drafting if the plan is wrong
• Use /story-long-write for longer work