Adversarial multi-view fiction review (full/lean/solo)
Review novel chapters across structure, characters, prose and setting with a 4-agent pipeline and severity-ranked S1–S4 findings — written in Chinese
- What
- Review novel chapters across structure, characters, prose and setting with a 4-agent pipeline and severity-ranked S1–S4 findings — written in Chinese
- Cost
- Free
- Needs
- the oh-story-claudecode reviewer agents deployed in my project by /story-setup, for the full and lean multi-agent modes (this version expects agents bundle version 36; a version mismatch is only flagged in a notice, never blocking); Node.js, for the skill's bundled read-only pre-check scripts. Without the deployed agents the skill degrades to solo mode automatically, and it falls back to its built-in rubric when its reference files cannot be read. Installing the skill itself is only files, and nothing runs during installation.
- Install
- Copy the installer prompt below into your Muse — your agent does the rest.
Curated by Skill Harbor — @zenstory-ai's review coordinator from the `oh-story-claudecode` writing toolkit: an iron rule that review finds problems rather than validating correctness. `/story-review full` spawns 4 reviewer agents (story-architect, character-designer, narrative-writer, consistency-checker), `lean` spawns two, `solo` runs one — with automatic degradation (missing/malformed agents, spawn failures, runtime without sub-agent support) and an embedded fallback rubric pack when reference files are unreadable. Every reviewer emits findings in a unified S1–S4 schema (severity, category, location, evidence, issue, fix), reports are written for the author (must-fix / should-fix / can-skip, quoted evidence, reader-impact reasoning, no internal jargon), and a fixed technical-note line records the actual mode, fallback and rubric used. Honest caveats: the skill is written in Chinese (style resolution follows the Chinese copy-layout guide); targeted at web-fiction platforms (Fanqie, Qidian, Zhihu rubrics built in); `story-review` never edits prose, settings or outlines — it only reports; author-habit memory may explain intent but never lowers rubric severity. By @zenstory-ai, listed here with credit to its creator. MIT-licensed. Skill Harbor never reviews the code, review it yourself before use. Discovered via skills.sh.
Version:
Quick answer
Adversarial multi-view fiction review (full/lean/solo) is a free Muse skill that reviews novel chapters across structure, characters, prose, and setting with a 4-agent pipeline and severity-ranked S1–S4 findings. The full mode spawns 4 reviewer agents, lean spawns two, and solo runs one, with automatic degradation and a fallback rubric pack when reference files are unreadable.
Install
Copy the install package below, then paste it into Muse🧪 Install ran in our test sandbox · Nov 18, 2026📌 Pinned version: commit 7408014Step 0, before installing anything: read this install prompt, summarize in a few lines what this build does and what installing it involves, then ask me to confirm before you proceed. Prerequisites: the oh-story-claudecode reviewer agents deployed in my project by /story-setup, for the full and lean multi-agent modes (this version expects agents bundle version 36; a version mismatch is only flagged in a notice, never blocking); Node.js, for the skill's bundled read-only pre-check scripts. Without the deployed agents the skill degrades to solo mode automatically, and it falls back to its built-in rubric when its reference files cannot be read. Installing the skill itself is only files, and nothing runs during installation. Install "Adversarial multi-view fiction review (full/lean/solo)" for me. It coordinates an adversarial review of a fiction text whose only job is to find problems — structure, characters, prose, setting — with evidence quoted from the text and actionable fixes: a Phase 0 pre-check first decides the mode, in full mode it spawns all 4 deployed reviewer agents in parallel (story-architect, character-designer, narrative-writer, consistency-checker), in lean mode 2 of them (story-architect, consistency-checker), then merges everything into one consolidated report using a single findings schema with severities S1–S4 across all modes; in solo mode it reviews on its own. It is triggered by /story-review (also /审查 and phrases like "review this"), by @zenstory-ai, MIT-licensed. Repository: https://raw.githubusercontent.com/zenstory-ai/oh-story-claudecode/74080143dd1ac9a39a930fafd99a0dbbe61b946a/skills/story-review/SKILL.md (the skill is the whole skills/story-review/ folder at that commit: SKILL.md plus its references/ subfolder, including the rubrics/ platform files, and its scripts/ subfolder) Pinned version: install the code at commit 74080143dd1ac9a39a930fafd99a0dbbe61b946a exactly (Skill Harbor tested this version). Do not substitute a newer version. If this commit is unreachable, stop and tell me. Updated October 18, 2026: Skill Harbor regenerated this install prompt from the current upstream file. The skill stays at version 1.1.1, but its reference set and review rules changed, so the whole folder must be reinstalled, not SKILL.md alone. Four reference files were removed from the folder (references/review-quality.md, references/plot-core-methods.md, references/character-relations.md and references/dialogue-mastery.md); the plot-loop, climax-building and relationship-progress checks they carried now live in the benchmark summary the coordinator builds for every review. Rubric loading was restructured: the generic rubric (references/quality-rubric.md) is now always loaded, and one platform rubric is layered on top of it by text length and platform (long-form: references/rubrics/qidian.md or references/rubrics/fanqie.md; short-form: references/rubrics/zhihu.md), with short-form reviews also checking format, quote style and lead-in against the platform tone in the book's 设定.md. The default review scope is sharper (in a git project, the changed prose, setting and outline files; otherwise the most recently modified chapter of a long-form book, or 正文.md for a short-form one, with the report stating what was reviewed), and tracking maintenance now applies only to long-form projects that carry a 追踪/_tracking-state.json file and only in full/lean mode — solo never modifies the tracking folder. Correction: the previous prompt said this version expects agents bundle version 34; that is no longer true — the current file expects agents bundle version 36 (a mismatch is still only flagged in a notice, never blocking). 1. Fetch the whole skills/story-review/ folder at the pinned commit, SKILL.md and every file under references/ (including references/rubrics/) and scripts/, into a temporary folder, then compare SKILL.md with the fingerprint from Skill Harbor's review of 2026-10-18: - SKILL.md: SHA-256 d6dc807b06ab9d6ef06a0532c6b2b5bd0da20d7fe1fc35505fdfb0cdad6558b3 If it does not match, it changed since the review: do not install it yet, tell me what you found and let me decide. 2. Safety check: read SKILL.md and its reference and script files and review them for anything suspicious (unexpected network calls, shell commands, credential harvesting); STOP on any red flag and tell me. Tell me in plain words what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Confirm these points from the files: - Its only job is to find problems: it reviews and reports, it does not fix the text. Its three bundled pre-check scripts (scripts/normalize-punctuation.js, scripts/check-ai-patterns.js, scripts/check-degeneration.js) are read-only checks, and the skill never modifies the manuscript, setting or outline files; its only writes at review time are the tracking files on long-form projects that carry a 追踪/_tracking-state.json file (full/lean mode only, at wrap-up) and the batch state file .story-review/state.md on batched reviews. - Modes are decided by a Phase 0 pre-check: full spawns all 4 reviewer agents (story-architect, character-designer, narrative-writer, consistency-checker), lean spawns story-architect and consistency-checker only, solo spawns none. Any missing or malformed agent file, an unavailable agent tool, a ZCode project (whose project custom agents cannot run), recursion inside a subagent, or a failed spawn degrades the review to solo automatically, with the reason recorded in the report's technical note; partial results from a failed spawn are never presented as a full/lean verdict. - Before spawning, it reads the agents_version marker in the project's .story-deployed file and compares it with the bundle version this skill expects, 36. A mismatch only adds a plain-language notice telling the author to rerun /story-setup and start a fresh session (or, if the project's bundle is above 36, to update oh-story-claudecode first rather than let an old local setup downgrade it); it never blocks and never downgrades on its own. - It only receives author preferences the caller already holds; it never queries or records author memory, and a standalone run loads no memory files. Preferences can help explain intent and organize the report, but they cannot lower a rubric severity, declare a factual conflict a non-issue, or skip a platform gate. Style resolution reuses the caller's style_resolution (deciding from the current request and the book's style only when invoked standalone) and hands the same ruling to every reviewer. - Rubrics: the generic rubric references/quality-rubric.md is loaded on every review, with one platform rubric layered on top by length and platform (long-form Qidian or Fanqie; short-form uses references/rubrics/zhihu.md). Reference files are resolved through the documented lookup order across the runtimes' skill folders; any file that cannot be read is replaced by the built-in fallback summaries in SKILL.md, and the report's technical note records whether each rubric came from a file or from the embedded fallback. - All modes use one findings schema with severities S1–S4 (the same scale as the long-form writing skills and the detectors); in the author-facing report the section headings read Must fix (S1–S2), Suggested fixes (S3) and Optional (S4), reviewer names and internal labels stay out of the body, and the last line is a fixed-format technical note (requested and actual mode, fallback, rubric source). - Nothing is sent anywhere during installation itself, and the skill connects to no network service when used: it reads local project files and runs its bundled local scripts. 3. When I say yes, install it as a skill: copy SKILL.md and all its files into the agent's skills directory, in a folder named "story-review", keeping the references/ and scripts/ subfolders as they are. 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (e.g. run /story-setup in my project and start a fresh session if I want the full/lean multi-agent modes with agents bundle version 36; have Node.js available for the pre-check scripts; the skill is written in Chinese, so work in Chinese when invoking it). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.
Saved to your recent installs. Find it anytime on /connect.
🔒 Security, provenance and what this touches
The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?
Verification certificate
Dated facts about what was checked for this listing, and how. This is not a safety certification.
- Pinned version
- 7408014 · 2026-10-18
- Sandbox install test
- Install ran in our test sandbox · Nov 18, 2026 · Observed during the install test: registry.npmjs.org
- Creator identity
- Not verified
- Fingerprint watch
- Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
- Install prompt analysis
- 2026-10-13 : No commands, network calls, packages, file writes or API keys detected.
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
Questions
How do I install a skill?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.