Guide · Agent skill
How to use the story-review agent skill for fiction review
Updated · written with AI from the project's own documentation and checked against our catalogue data
story-review is an agent skill that makes an AI assistant act as a fiction editor. You give it a chapter or a set of chapters, and it reports structural, character, dialogue, consistency and prose problems, each tied to a quote from your text. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what it needs, so you can run your first review in one sitting.
Story Review Editorzenstory-ai/oh-story-claudecode · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the story-review ZIP from this page and unzip it into ~/.claude/skills/, or into a project's .claude/skills/ folder.
- 2Or upload it in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the skill there.
- 3Start a new session. Open a new session so the agent loads the skill.
- 4Ask for a review. Name the chapter or file and say 審查一下 or use /story-review, optionally adding full, lean or solo.
- 5Read the findings. Work through must-fix items first, check each quoted passage, then decide on suggestions.
What an agent skill is and what story-review makes the agent do
An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a certain kind of task. Once the skill is installed, the agent can load these instructions when a request matches the skill's trigger.
With story-review loaded, the agent acts as a fiction editor. When you ask it to review a chapter, it checks structure, character motivation, dialogue, consistency with the story's established facts, and prose quality against written rubrics. It then reports problems ranked by severity, with quotes from your manuscript as evidence.
The skill is written to report and suggest, not to rewrite. It is meant to find problems and explain them, leaving the changes to you.
- Structure, character motivation, dialogue, consistency and prose quality are the review areas.
- Each problem is ranked by severity and quotes the text it refers to.
- It can run several sub-reviewer agents in parallel, or do a single-pass review on its own.
How to install the story-review skill
Option 1: download the ZIP from this page and unzip it into ~/.claude/skills/ so the skill is available across your projects. Option 2: unzip it into a project's .claude/skills/ folder so it applies only to that project.
Option 3: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill there.
After installing, start a new session so the agent picks up the skill. The skill is available under the MIT licence.
- Global install: ~/.claude/skills/
- Per-project install: the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload
How to trigger a review with natural requests
You can start a review with the slash command /story-review, or with the Chinese triggers /审查, 「审查一下」 or 「帮我审一下」. Adding a mode word after the command chooses how the review runs.
If you do not name a mode, the skill uses full mode. Full mode tries to run all four sub-reviewer agents in parallel. Lean mode uses two of them, story-architect and consistency-checker. Solo mode runs no sub-agents and the current session does the review alone.
- /story-review or /story-review full: full mode, four sub-reviewers when available
- /story-review lean: story-architect and consistency-checker
- /story-review solo: single-session review with no sub-agents
- Natural phrasing such as 「审查一下」 or 「帮我审一下」 also triggers the skill
/story-review full
/story-review lean
/story-review soloUnderstand the review report
Tell the agent which chapter or file to review. If you name nothing, the skill reviews the files it finds changed in a git project, or the most recently changed chapter in a long-form project. The report opens by stating what was reviewed.
Findings are grouped for the author. Problems marked as must-fix are the most serious. Suggested fixes are less urgent, and optional fixes are minor. Each entry gives the chapter and the quoted passage, explains the effect on the reader or story, and suggests a fix you can accept or ignore. Where something needs your decision, the report asks you directly.
The last line of the report is a technical note that records the requested mode, the mode actually used, and any fallback.
- Must fix: the most serious problems, such as clear factual conflicts
- Suggested fix: problems that weaken the reading experience
- Optional: minor issues you may choose to skip
- Each finding cites the chapter and quoted text
What the skill needs and safety notes
The skill needs an agent runtime that reads SKILL.md skills, such as Claude Code, the Claude apps, the Agent SDK or a similar tool. Sub-agent definitions from the oh-story-claudecode setup are optional and are needed only for full or lean mode. Solo mode works without them.
Platform rubric files for fanqie, qidian and zhihu are optional. If they are missing or unreadable, the skill uses built-in fallback rubrics. The listing states that the skill does not use the network. It also states that it does not run scripts. The skill text does mention helper scripts for punctuation and AI-pattern checks, and a tracking check. Those files are not in the listed file set, so look at what is in your copy before relying on them.
If a sub-agent cannot start, or a required agent definition is missing or broken, the skill falls back to solo mode and says so in the technical note.
- Required: an agent runtime that reads SKILL.md skills
- Optional: sub-agent definitions for full or lean mode
- Optional: platform rubric files, with built-in fallbacks
- Network: not used according to the listing
- Scripts: the listing says none are run; check your copy for helper scripts
Tips for getting useful reviews
Give the skill a narrow scope. A single chapter or a small batch gives more precise findings than a whole book. For long manuscripts, the skill splits the work into batches and keeps a progress file in a .story-review/ folder so later batches can pick up open items.
Treat the rubrics as checklists, not verdicts. A stylistic choice you made on purpose may be flagged, and you can overrule it. Keep your own judgement for literary quality.
- Review one chapter or a small group at a time
- Keep your setting and outline files up to date so consistency checks work
- Read every quote before accepting a fix
- Use solo mode when you want a faster pass without sub-agents
What people use it for
- A web-novel author doing a structured second read of a chapter before publishing it.
- A serial writer checking whether a character's behaviour matches facts set in earlier chapters.
- An author who wants suggestions backed by quoted evidence rather than a rewritten draft.
- A writer running a quick solo-mode pass on a short story without sub-agents.
Frequently asked questions
How do I install the story-review skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
Is story-review free to use?
The skill is published under the MIT licence, as stated on its listing. This page does not list any price, so check the listing for current terms of use.
What do I need to run story-review?
You need an agent runtime that reads SKILL.md skills, such as Claude Code, the Claude apps or the Agent SDK. Sub-agent definitions from the oh-story-claudecode setup are needed only for full or lean mode, and solo mode works without them.
Does story-review rewrite my manuscript?
No. The skill is written to report problems and suggest fixes, not to change your text silently. You decide what to revise.
What if the sub-agents are missing or fail?
The skill falls back to solo mode and records the fallback in its technical note. It does not treat partial sub-agent results as a full or lean review.
How does story-review compare with a human editor?
The skill gives a structured, quote-backed second read, which is useful before publishing or when checking consistency across chapters. It is not a final judgement of literary quality, because its rubrics are checklists and can flag intentional stylistic choices.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
story-review suits web-novel authors and serial writers who want a structured second read with quoted evidence, especially for consistency across long projects. It is a poor fit if you want a final verdict on literary quality or an agent that rewrites your manuscript automatically, so pick a human editor or a rewriting tool for those jobs.
Download the skill