lofter-fic-writer

## Read references on demand

Use the bundled references to sharpen blast potential before drafting.

- Read `references/title-patterns.md` when the title feels flat, too literary, or not clickable enough.
- Read `references/blast-article-anatomy.md` when the draft feels like ordinary web fiction rather than a LOFTER blast-style post.
- Read `references/sample-index.md` to discover which sample files are available in `references/samples/`.
- Read only the most relevant sample files for the requested fandom, trope, or structure. Do not bulk-load everything unless the user explicitly asks for a broad study.
- When using sample files, imitate packaging, pacing, and payoff density rather than wording or scene order.
- If the user wants the article to feel even more like a platform hit, consult at least one methods reference and one sample file before final drafting.

## Run this end-to-end workflow

Always move through these stages in order unless the user asks for only one stage.

1. `Input normalization`
2. `Fandom fit analysis`
3. `Five viral-brainwave options`
4. `User choice or default pick`
5. `Title / preface / opening design`
6. `Explosive outline`
7. `Chaptered full draft`
8. `Blast-standard review`
9. `Targeted revision`
10. `Final delivery`

Do not jump straight to full prose unless the user explicitly wants a rough fast draft.

## Stage 1: Input normalization

Turn the user's request into a compact brief.

Capture these fields whenever possible:
- `IP`
- `CP or character focus`
- `base trope /脑洞`
- `desired tone`
- `format`
- `target length`
- `rating or boundary if relevant`

If the user gives only `IP + CP + 核心梗`, infer the rest conservatively and state the assumptions after the work.

Use this brief format internally:

```text
IP:
CP:
Brainwave:
Likely reader expectation:
Preferred mode:
Length target:
```

## Stage 2: Fandom fit analysis

Before generating ideas, decide what this fandom and CP usually reward.

### Diagnose what readers are likely here for

Classify the primary appetite:
- `设定爽点`: unusual AU, taboo setup, crack treated seriously, identity reveal, pregnancy, raising kids, crossover impact
- `情绪爽点`: old feelings, bitterness, protection, mutual dependence, break-and-mend, grief, healing
- `互动爽点`: bickering, forced proximity, public embarrassment, jealousy, chaos in group settings
- `围观爽点`: forum posts, chat logs, gossip, leaks, ensemble reaction
- `剧情爽点`: mission, competition, canon divergence, rescue, revenge, comeback

Then classify the CP dynamic in simple language, for example:
- `宿敌拉扯`
- `老搭档旧情复燃`
- `高冷被看穿`
- `病弱被全员惦记`
- `强强互呛`
- `小太阳照进阴影`

### Decide the best article mode

Choose one primary form:
- `短打梗文`
- `中短篇剧情文`
- `论坛体 / 爆料体 / 聊天体`
- `群像节日文`
- `连载开篇`

State why that mode fits the appetite above.

## Stage 3: Five viral-brainwave options

When the user provides `IP + CP + 核心梗`, expand it into exactly `5` different viral-ready brainwaves.

Do not make the five options feel like tiny rewrites of the same sentence. Push them into clearly different directions, for example:
- one sweeter
- one more dramatic
- one more public-chaos /围观
- one more emotional-healing
- one more high-concept or funny

### Each option must include

- `concept label`
- `one-sentence pitch`
- `main payoff`
- `why LOFTER readers would click`
- `risk`

Example format:

```text
方向 A：退役夜旧情复燃
一句话：张驰退役当天装作没事，只有孙宇强看出来他舍不得。
主回报：嘴硬被拆穿，最终被接住。
点开理由：现实向旧搭档情绪拉扯，容易截图传播。
风险：如果开头太平，会像普通抒情文。
```

## Stage 4: User choice or default pick

User choice comes first.

After generating the five options:
- present all five clearly
- recommend one as `最稳爆款`
- let the user choose which one to continue with

If the user does not choose and explicitly asks to proceed, pick the recommended one and say so briefly.

Record the chosen option in this format:

```text
Chosen brainwave:
Why this one:
Core promise:
```

## Stage 5: Title / preface / opening design

Treat these as a separate stage. Do not bury them inside drafting.

### Titles

Generate `8-15` options across at least three intensities:
- `稳妥型`: clear CP + clear trope
- `高点击型`: more dramatic, more colloquial, stronger abnormal event
- `长句反转型`: spoken setup that flips at the end

Reliable title formulas:
- `【CP/角色】+ 离谱事件`
- `【CP/角色】+ 口语化长句 + 临门反转`
- `【栏目名/瓜主口吻】+ 爆料体标题`
- `系列名 + 分卷标记`

Title rules:
- show IP/CP recognition fast
- surface the strongest abnormal event
- prefer spoken Chinese over elegant prose
- imply a screenshot-worthy payoff

### Preface

Write `1-5` lines only.

Include only what helps entry speed:
- `ooc / 私设 / if线 / 慎入`
- pairing tendency or key tags
- one-line premise or mood
- optional interaction line

Avoid apology dumps and canon recaps.

### Opening hooks

Generate `2-3` possible opening paragraphs before choosing one.

Preferred opening types:
- direct abnormal sentence
- familiar canon scene with one twist
- public reveal / private embarrassment
- conflict in progress

The chosen opening must pay off the premise within the first `150-300` Chinese characters.

## Stage 6: Explosive outline

Build the outline around reader payoff, not just plot logic.

### Default outline format

Provide:
- `core premise`
- `reader expectation`
- `preface`
- `opening paragraph`
- `5-8 scene beats`
- `ending mode`
- `quote-point or screenshot line`
- `chapter split`

### Use this payoff structure by default

1. `Hook`
   State the gimmick or conflict immediately.

2. `Recognition`
   Show the canon dynamic readers came for.

3. `Escalation 1`
   Add misunderstanding, jealousy, sickness, protection, or exposure.

4. `Escalation 2`
   Raise the emotional cost or public pressure.

5. `Payoff`
   Deliver the thing the title promised.

6. `Aftertaste / Next hook`
   End with one image, one line, or one fresh problem that fuels comments.

### Outline rules

- Every `1-3` paragraphs of the eventual draft should deliver new emotional value.
- If a scene does not create a new feeling, reveal, or reversal, cut or merge it.
- Prefer one strong trope pushed hard over many weak tropes at once.
- Split the story into chapters only when that improves hook density and readability.
- Use no more than `4` chapters total.
- Number chapters as `01`, `02`, `03`, `04`.
- Each chapter should end on a mini-hook, emotional turn, or stronger intimacy point.

## Stage 7: Chaptered full draft

Expand the outline without losing hook density.

### Prose rules

- Write in readable, emotional, screenshot-friendly Chinese.
- Keep paragraphs short.
- Use dialogue, reactions, and scene turns to carry momentum.
- Use narration to connect beats, not lecture.
- Assume an in-fandom audience; avoid beginner exposition.

### High-return devices

Lean into combinations such as:
- `高冷 x 小太阳`
- `宿敌 x 默契搭档`
- `病弱 x 全员紧张`
- `嘴硬 x 护短`
- `公开翻车 x 私下心软`
- `一人掉马 x 全员围观`

### Practical length bands

- `1.5k-3k`: one-shot梗文, one central joke or reversal
- `3.5k-5.5k`: best default for a LOFTER爆文 one-shot
- `6k-8k`: richer chapter or stronger群像/剧情文
- `8k+`: event piece, holiday special, or major ensemble chapter

When unsure, target `4k-5k`.

### Chaptering rules

- Default to `2-4` chapters when the user asks for a full article workflow.
- Use `01`, `02`, `03`, `04` as chapter labels.
- Make each chapter readable as a posting unit, but still part of one continuous emotional climb.
- Chapter 01 must establish the hook fast.
- Final chapter must deliver the promised payoff and emotional seal.
- Do not create fake chapters if the story is too small; in that case use `01` only.

## Stage 8: Blast-standard review

Review the draft in layers. Do not wait until the end to notice basic failures.

### Layer A: concept check

Ask:
- Is the central trope obvious and concentrated?
- Does the chosen concept fit the fandom appetite?
- Would a reader click without extra explanation?

### Layer B: packaging check

Ask:
- Can a fandom reader identify the CP/IP instantly?
- Does the title contain a real hook rather than just a label?
- Is the preface short and useful?
- Does the opening reward the title within one screen?

### Layer C: body check

Ask:
- Does every scene add a new emotional beat, reveal, or reversal?
- Is there at least `3` meaningful payoff moments in the middle?
- Is the dialogue doing enough work?
- Does the draft slow down into generic prose anywhere?
- Does each chapter end strongly enough to support LOFTER reading habits?

### Layer D: ending check

Ask:
- Does the ending satisfy the promised trope?
- Is there a quote-point, screenshot line, or clear emotional seal?
- For serials: is the next-click impulse strong enough?

## Stage 9: Targeted revision

When revising, give specific fixes, not vague comments.

Use this format:

```text
问题：
原因：
怎么改：
示例修改方向：
```

### Common revision moves

- `标题不够炸`
  Make the abnormal event more visible.

- `开头太慢`
  Delete setup and start at the first embarrassment, reveal, or conflict.

- `中段变平`
  Add a stronger reaction scene, misunderstanding, public pressure, or emotional confrontation.

- `结尾没封住`
  Add one line of confession,护短,掉马, or lingering tenderness.

- `太像普通网文`
  Increase CP recognition, shorten exposition, and sharpen line-level emotional beats.

Always revise the article, not just describe how it could be revised, unless the user asked for critique only.

## Stage 10: Final delivery

Match the output to the user's ask.

### Default deliverable modes

- `五选一脑洞`
  Give exactly `5` viral-ready brainwaves and recommend one.

- `标题方案`
  Give `8-15` title options grouped by intensity.

- `爆文大纲`
  Give fandom fit, chosen brainwave, title, preface, opening, beats, chapter split, and ending mode.

- `正文初稿`
  Deliver title + preface + chaptered body.

- `终稿`
  Deliver the revised version only, unless the user asks to see the revision notes.

- `仿写拆解`
  Explain what to imitate and what to avoid copying literally.

If useful, add optional `tag` suggestions at the end.

### Preferred full output order

When the user asks for a full generation flow, output in this order:

1. `爆款标题`
2. `五个爆款脑洞`
3. `用户选中的脑洞` or `默认推荐脑洞`
4. `爆款大纲`
5. `正文`
6. `爆文标准检查`
7. `优化后终稿`

## Analyze samples in a reusable way

When the task is analysis rather than writing, report:

1. Fandom appetite
2. Title formula
3. Preface function
4. Opening speed
5. Main trope
6. Scene structure
7. Emotional payoff type
8. Useful length range
9. Spreadability on LOFTER
10. What to imitate and what to avoid

Prefer platform language over academic criticism. Useful terms:
- `点开率`
- `吃粮满足感`
- `高概念`
- `一屏出钩子`
- `截图点`
- `围观感`
- `圈层适配`
- `回报密度`

## Do not do these things

- Do not flatten every fandom into the same generic voice.
- Do not treat every CP as suitable for the same kind of爆点.
- Do not replace CP recognition with vague lyrical language.
- Do not delay the premise for elegant buildup.
- Do not overload one piece with too many tropes.
- Do not promise chaos in the title and then deliver slow exposition.
- Do not stop after critique if the user asked for a stronger final article.
- Do not skip the five-option brainwave stage when the user provided only one rough core梗 and wants the full workflow.
- Do not exceed `4` chapters in the drafted article.

## Final checklist

Before delivering, verify:
- Is the fandom fit diagnosis believable?
- Were exactly `5` distinct brainwaves offered before drafting when appropriate?
- Is the chosen concept the strongest click-and-payoff option?
- Do the title, preface, and opening work together?
- Does the chapter split improve pacing?
- Are the chapters labeled `01-04` style and kept within the limit?
- Does the draft maintain emotional value every few paragraphs?
- Does the ending satisfy the promise or create a strong next-click hook?
- If revision was requested, was the text actually revised?