Not just finding faults, but diagnosing the manuscript from thereader retentionperspective, giving suggestions that can be directly revised. This is the core that distinguishes this plugin from general writing tools.
When to use
- Review one chapter / one volume / the whole book
- Check plot holes, foreshadowing breakpoints, character OOC
- Diagnose why follow-read rate/completion rate drops
- Pre-publication checkup
Six-dimensional diagnosis
1. Hook diagnosis (retention)
- Do the opening 3 sentences establish a crisis/contrast?
- Does each chapter have an effective chapter-end hook? Mark chapters with "flat endings" (adjust
references/hook-library.md） - Is a strong hook pressed in before the paywall node?
2. Satisfying-moment diagnosis (follow-read)
- Does the density of satisfying moments match the platform? (Fanqie every chapter, Qidian every 3 chapters, adjust
references/platform-profiles.md） - Mark consecutive chapters with "satisfying-moment gaps"
- Are satisfying moments repeated in place without escalation? (adjust
references/trope-library.mdescalation chain) - Is the four-beat face-slapping complete (especially whether the "onlookers" beat is missing)
3. Foreshadowing diagnosis (prevent bad endings)
- List all planted foreshadowing → mark: paid off / pending payoff / suspected forgotten
- Mark "holes dug but not filled" and "settings that suddenly appear without setup"
4. Character-setting diagnosis (consistency)
- Are characters' words and actions consistent throughout, is there OOC
- Does the protagonist have subjective agency (or is he led along by the cheat)
- Is the villain's strength enough to support the protagonist's victory
5. Pacing diagnosis (sense of breathing)
- Is it tense all the time or flat all the time?
- After battle/face-slapping, is there relief, and after relief, is tension pulled tight again
6. Human-flavor/loss-of-control diagnosis (does it read like a real person wrote it)
- Are the emotions too "average" (everyone reacts reasonably, no one truly loses control/is obsessive/is in pain)?
- Is the conflict too "clean" (misunderstanding → explanation → resolution, lacking "rolling dirtier and dirtier")?
- Are the sentences too "smooth" (no variation in length, no pauses, no incomplete expressions)?
- Are the characters too "normal" (no irrationality/obsessiveness/gambling of "knowing it's wrong but doing it anyway")?
- Is it too "neutral" (no bias, no sense of personal agenda)?
- One-sentence standard: does it read like amanualoremotional record? Too much like a manual → use
humanskill to add loss of control (adjustreferences/human-texture.md）。 - Wantquantificationto "AI-flavor index X/10 + which sentences are deducted" → use
aidetect(metaphor density/consecutive same sentence patterns/sentence-length fluctuation/paragraph distribution/emotion-word repetition, seereferences/ai-detector.md）。
Output: diagnostic report
【整体评分】留人 / 追读 / 连贯 各打分 + 一句话结论
【继续率预测】前三章：冲突/悬念/人物记忆点/反差 → 读者会不会追下去（高/中/低 + 为什么）
【致命问题】最影响数据的 1–3 点（按优先级）
【钩子】哪几章收尾平淡 → 怎么改
【爽点】哪段断档 / 没升级 → 怎么补
【伏笔】待回收清单 + 疑似遗忘的坑
【人设】OOC 或强度问题
【节奏】哪里需要松/紧
【修改优先级】先改哪个见效最快
Principles
- Diagnosis must bespecific to chapter/paragraph, no empty talk
- Every problem must be given anactionable fix, not just pointing out the problem
- Priority ordering: fix what most affects reader retention first
Next step
According to the report → use expand to rewrite the problem chapter,shuangdian add the payoff,hook fix the hook.
This skill isreader retentionperspective (hook/payoff/binge-read). Toauthor backendperspective deeply check settings for no contradictions (behavior/causality/resources/identity/injury/combat power/timeline) → use
continuity; to get a quantified AI-flavor score → useaidetect。