@wji.ai/cli 0.4.0 → 0.5.0

package/skill/short-drama/reference/story-mystery.md

@@ -0,0 +1,127 @@
+# Story genre `mystery` — 悬疑
+
+推理 · 刑侦 · 悬疑复仇 · 惊悚. In 短剧 the pure puzzle is rare and hard to sell (红果's 悬疑推理
+category drew 200+ titles, very few past 5,000万 heat); the hits **blend suspense with an emotional
+payoff** — 《引她入室》 is a family revenge story in a suspense frame. So the default here is
+悬疑 + 复仇 / 身份 爽点, with clue-solving as the way the 智商碾压 or the 身份揭露 is delivered.
+Shared machinery: `story.md`.
+
+## 1. The promise — curiosity, dread, the click
+
+- **Three pleasures**: curiosity (what happened, who did it), dread (what will happen, when), and
+  the click — the answer was in plain view all along.
+- **Who knows what** (McKee): mystery (the characters know more → curiosity), suspense (both know
+  the same → worry), dramatic irony (the audience knows more → dread).
+- **Hitchcock's bomb under the table**: surprise buys 15 seconds; showing the bomb and the clock
+  buys 15 minutes. **Keep the central whodunit a mystery, but run most episodes as suspense or
+  irony** — the audience sees the threat and the clock, the character does not. Irony delivers
+  tension in 60 seconds; a pure puzzle needs setup time the format does not have.
+
+## 2. The series — one question, layered secrets
+
+- **Play the cards early.** Plant every 梗 and 伏笔 in roughly the first 30 episodes and use the
+  rest to decode them. The first ten are the golden window: the main line, the core hook, and
+  "what kind of story this is".
+- **A central question over everything, small questions that last an episode or two** ("a ball of
+  questions rolling"). Each answer exposes a deeper question.
+- **逻辑闭环**: every step from the crime to the solution is explained, and every detail the
+  investigator uses was set up earlier.
+- **Case-of-the-week plus a through-line** works (《朱雀堂》: each case closes, the main-line
+  伏笔 accumulate); so does a jigsaw of timelines (《漫长的季节》, the two lines merging at the end of
+  episode 1 to pay off the device early).
+
+| phase | what |
+| --- | --- |
+| golden (~E1–10) | the inciting crime or secret in E1; the **central question stated by E3**; the investigator, the victim and all major suspects — **the culprit included** — on screen; a first small question answered by E3; a layer-2 reversal at the paywall |
+| escalation | layered secrets; false suspects rise and fall; a **midpoint twist at 40–60 %** that reframes known clues rather than adding new ones |
+| climax | the central truth, **assembled from planted clues**, at 80 % or later; the justice or revenge payoff; 1–3 episodes of aftermath |
+
+**Tracked in the outline** (gate `story-mystery`):
+- `outline.mysteryMode`: `whodunit` (default) · `inverted` (the audience knows the culprit — the
+  question is how they will be exposed) · `thriller` (a threat, not a puzzle: no culprit rule).
+- `outline.questions[]: { id, layer: 1 central / 2 major / 3 local, question, openEp, answerEp, clues[] }`.
+- `outline.clues[]: { id, kind: true / red-herring, what, plantEp, prop?, echoEps[], dismissedEp? }`.
+- `characters[].mysteryRole` (`investigator` `culprit` `accomplice` `suspect` `unreliable-ally`
+  `victim` `witness`) and `firstEp`.
+
+**The rules** (researched defaults, waivable on the user's request):
+
+| rule | default | why |
+| --- | --- | --- |
+| `central-question` | exactly one layer-1 question, opened by E3, answered at ≥ 80 % (runs ≥ 10) | the golden window; the truth in the climax |
+| `culprit-early` | exactly one culprit, first on screen by E5 / the golden phase (never after 30 %) | Knox 1, Van Dine: the culprit appears early and plays a real part |
+| `clue-count` | an answered question has ≥ 3 true clues (layer 1), ≥ 2 (layer 2), ≥ 1 (layer 3) | fair play: every clue to the solution is shown first |
+| `clue-lead` | a clue is planted ≥ 1 episode before a local answer, ≥ 3 before a major one, ≥ max(10, 15 %) before the central one; one central clue in the golden phase | too close and they see it coming; too far and they forget |
+| `clue-echo` | a clue planted more than 20 episodes before its payoff is echoed within 15 episodes of it | fragmented viewers need a reminder |
+| `red-herrings` | fewer red herrings than true clues; each dismissed before the truth | too many and the audience feels tricked, not tested |
+| `open-questions` | at most 6 open at any episode | the fragment viewer is lost past that |
+| `clue-on-screen` (script) | the beat that plants a clue is marked `clue: "K1"` in its plant episode | a clue is something the camera shows, never narration |
+
+`story-advice` also notes a whodunit with fewer than 3 or more than 5 viable suspects and 反转
+cliffs above 35 % (twist fatigue).
+
+## 3. The episode
+
+- Conflict or a reversal within 15 s, the story moving by 30 s, a suspense beat in the last 10 s;
+  95 of 100 endings leave a hook. 红果 asks for three hooks a minute in the first 30 % of a 拉新
+  drama.
+- **Answer one, open one or two.** Each episode closes or partly answers a small question and
+  opens or escalates another. Quality over quantity; alternate 紧张 and 松口气.
+- **Hooks** (红果 tutorial): an unfinished accusation; an identity hint ("你真以为我不知道你是
+  谁？"); a deadline; the cut at the instant emotion erupts; an unexpected figure; **a familiar face
+  flashing across a CCTV frame**.
+- **Cliffs** by hook type: 身份 (the anonymous helper is…), 危机 (the bomb clock), 道具 (a clue object
+  where it should not be), 反转 (an accepted fact overturned), 结果 (the body, the test result),
+  情感 (the ally is lying).
+
+## 4. Characters and suspicion
+
+- **The investigator** has a distinctive skill or persona and solves the case through their own
+  reasoning on shown clues; in 悬疑复仇 it is often the wronged protagonist.
+- **The culprit** appears early, is someone the audience cares about, is not obvious, and their
+  thoughts are **never shown** (no inner voice from the culprit before the truth). One mastermind;
+  helpers are accomplices.
+- **3–5 viable suspects**, each with motive, means and opportunity on screen. Rotate the prime
+  suspect ("前一刻认定的凶手，下一刻可能因新证据被推翻"). Give the real culprit one fair moment of
+  suspicion, then clear them with an explanation that is plausible and false.
+- **The unreliable ally** — the helper with a hidden agenda — is the standard 反转 engine.
+
+## 5. Clue craft
+
+- **Fair play**: every clue leading to the solution is shown before the reveal; the investigator
+  uses nothing the audience has not seen; no supernatural agency or unknown poisons unless the
+  genre says so (Knox, Van Dine).
+- **Chekhov**: every prop, skill or line introduced is a promise — pay it off or cut it.
+- **Hide in plain sight**: plant the true clue inside a beat that is doing something else (a 爽点
+  slap, an argument), in the background of a wide shot, or with an innocent explanation (the ring's
+  "flaw" read as marital symbolism). Bury it fourth in a list. **Red herrings may get the
+  close-up; true clues usually should not at the plant.**
+- **Pay off with a close-up and a flashback insert**, so fragmented viewers reconnect it. On the
+  storyboard the plant is often an `evidence` cut in a wide, the payoff an `evidence` close-up.
+- **Name puns** (吴池 = 无耻, 郑萋 = 正妻) are cheap, legible foreshadowing that suits 短剧.
+
+## 6. Dialogue and tone
+
+Separate what each character **knows** from what they can **prove**; every interrogation question
+is a thrust, not a static Q&A. Subtext is want or fear expressed indirectly, carried by the body.
+**悬疑创作是克制的艺术**: concrete details, no overflowing emotion. Characters withhold for a
+reason of their own, never because the writer needs a delay; every lie told on screen can later
+be caught against something already shown. "笑着去讲悲伤的故事" hits harder than heavy lines.
+
+## 7. What fails
+
+强行反转 and 降智 characters; twists for show; the mystery box that never pays off; 高开低走 (a big
+opening that collapses); a clue the audience never saw, a late culprit, a supernatural solution;
+too many red herrings; relentless intensity with no character growth; too many characters too
+early.
+
+## 8. Sources (read as search summaries)
+
+腾讯 / 搜狐 on 逻辑闭环 and clue density; 新浪 (producer 陈璐); 流媒体网 on the 红果 悬疑 category, 古装
+悬疑 craft (《朱雀堂》) and 卡点; 腾讯 / CSDN "all cards in the first 30 episodes"; 红果 tutorial 06
+(hooks) and 编剧第一课 (3 hooks a minute); 搜狐 (张静思); 电影界 / 百度百科 on 《引她入室》; 网易 / 知乎 on
+《漫长的季节》; 界面 (潘依然, 《隐秘的角落》); 知乎 (点灯) on restraint and 高开低走; DataEye 2025; 澎湃 on
+红果 2025; Hitchcock / Truffaut and Bordwell on the bomb; Knox's commandments; Van Dine's rules;
+McKee on mystery vs suspense; StudioBinder and iwrity on Chekhov's gun; Writer's Digest and The
+Write Practice on clues and red herrings; nailyournovel on the ball of questions; InkTip and
+Filmustage on vertical drama; interrogation craft (StudioBinder, Bang2write); 文汇报 on 全程高能.

package/skill/short-drama/reference/story-palace.md

@@ -0,0 +1,103 @@
+# Story genre `palace` — 宫斗 / 权谋
+
+宫斗 · 宅斗 · 嫡庶之争 · 夺嫡 · 权谋复仇 · 女帝 / 女相 大女主. It pairs with the `costume-dark` or
+`costume-idol` look and the `古装` register. Shared machinery: `story.md`.
+
+## 1. The promise — winning with the mind
+
+- **智性爽**: watching the heroine read the room, set a trap and flip the verdict. In 琅琊榜 "most
+  strategies rely on predicting how another person will react"; 《冒姓琅琊》 (2025) sold "知识即武器的智性爽感".
+  Revenge rewards a *power* reversal; 宫斗 rewards a *mind* reversal — the payoff is "she saw it
+  coming", not "she turned out to be rich".
+- **升级**: the rank ladder is the progress bar (延禧攻略: becoming stronger alongside her).
+- **道德资本**: the winner must deserve it — "I didn't want to fight; you forced me". Losers lack skill
+  or are exposed doing evil.
+- **A tragic undertow** (optional): 甄嬛 won at the cost of love; 庆余年 is "a 爽文 shell over a cruel
+  truth" — in 短剧, one bittersweet beat near the end.
+
+## 2. The series — schemes, a ladder, factions
+
+| act | ~episodes (of 80) | function |
+| --- | --- | --- |
+| 入局 | 1–10 | low-rank entry (选秀, 庶女, 替身, 重生); the first humiliation and small counter-win; the first rank gain; the antagonist's faction in action; the 卡一 cliff (E8–12) is **a scheme near-exposure** |
+| 立足 | 11–30 | scheme rounds of 3–6 episodes; allies, promotions, a patron; **the antagonist wins at least once** |
+| 低谷 | 30–50 | a big frame-up succeeds (冷宫, demotion, 禁足, a lost child); an ally betrays; she starts the long plant |
+| 反攻 | 50–70 | intel gathered, the antagonist's servants turned; factions realign; she rises faster |
+| 夺权 | 70–end | **the master counter-scheme planted in Acts II–III pays off**; the final verdict, the top rank or the regency |
+
+- **Long plants are the backbone**: 甄嬛传 plants "得一人心" in episode 1 and pays it off dozens later;
+  梅长苏's twelve-year setup unfolds as numbered moves (北燕, the 兰园藏尸案, 南楚); the 滴血验亲 counter-trap
+  begins the empress faction's collapse — **one scheme changes the factions**.
+- **Each scheme across 3–6 episodes**: 埋 (a prop or intel in passing) → 发 (the accusation; a 危机
+  hook) → 危 (the verdict seems certain; a cliff) → 反 (the counter from what was planted; 智商碾压, a 反转
+  hook) → 罚 / 升 (the antagonist punished, her rank changes). **Overlap the arcs**: scheme B is planted
+  during scheme A's reversal.
+- **Two modes of irony**, about half and half: the audience sees the poison go in (will she notice?),
+  or the heroine's plan is withheld (the reveal is the payoff). Too much withholding feels like a cheat.
+
+**Tracked in the outline** (gate `story-palace`): `outline.rankLadder: { track, levels[] low → high }`,
+`outline.rankEvents[]: { char, ep, to }`, `outline.schemes[]: { id, schemer, target, stakes: minor /
+major / master, setupEp, revealEp, outcome: success / countered / backfired, counteredBy: { type:
+intel / prop / ally / counter-trap / authority, ref, plantEp }, flaw, justice? }`,
+`outline.factions[]`, `outline.allegiance[]: { char, ep, from, to, motive, foreshadowEp }`.
+
+| rule | default | why |
+| --- | --- | --- |
+| `scheme-setup` | ≥ 2 / 5 / 20 episodes between setup and reveal (minor / major / master, the master scaled to the run); a master scheme planted in the first half | the long plant is the genre |
+| `no-teleported-evidence` | every counter is planted before the reveal; every countered scheme records the loser's flaw | no evidence out of nowhere; every loss has a cause |
+| `authority-not-rescue` | at most one scheme per phase countered by the emperor alone | the arbiter has an agenda |
+| `competent-antagonist` | an antagonist scheme succeeds against the heroine before the midpoint | a strong enough villain saves a story |
+| `rank-climb` | the first rank event by E12; ≥ N / 15 rises; no 15-episode gap; a low point at 30–60 %; the top rung in the last 10 % | the progress bar and its dip |
+| `faction-realignment` | 2–4 factions; every switch of sides has a motive and foreshadowing ≥ 5 episodes earlier | betrayals are foreshadowed |
+| `moral-capital` | the heroine only schemes against someone who schemed against her first (or records a justice goal) | 道德资本, and regulatory safety |
+
+## 3. The episode
+
+A cliff, a conflict and a reversal every episode (情绪未完、剧情刚翻新、人物状态临界值); the loop is
+冲突 → 行动 → 反转 → 爽点. A trigger or danger episode ends on 危机 / 反转; the reveal episode pays in
+智商碾压 / 打脸 / 复仇 / 身份揭露. Every accusation scene needs a witness chain: the maid → the 太医 → the
+evidence prop.
+
+## 4. Characters
+
+- **The heroine** goes naive → burned → strategist; with 重生 she starts as a strategist and the arc
+  is the *cost* (an ally, a child, love).
+- **The antagonist decides quality** — never a whetstone; every exit should feel regretful.
+- **The emperor** is the prize and the root of all evil (the 欢宜香); an arbiter with his own agenda,
+  never a free rescue. A fresh variant: the emperor as a two-way opponent (《枕边刃》).
+- **Allies play their own games** (安陵容 scoring a point in the chaos); **servants are pieces** — the
+  channel for intel and planted evidence; turning the antagonist's maid is the Act IV move.
+- **Every schemer has a motive** (宫斗 is driven by family, 宅斗 by individual will).
+
+## 5. Beats and freshness
+
+The kit: 麝香 in ointment, the 红麝香珠, 假孕, 滴血验亲, 巫蛊 and poison, the 私通 frame, 选秀, 请安立规矩,
+翻牌子, 冷宫, 禁足, 降位 / 晋封, 落水, 狸猫换太子, forged letters, scents, the 太医 testifying, 嫡庶, 分家, 夺嫡 and
+站队. **Fresh**: knowledge as the weapon (冒姓琅琊); comedy and workplace framing (《我在皇宫搞内卷》, 《谁在说朕
+坏话》); two strong leads with the emperor as the opponent; **reverse a stock trope** (the 滴血验亲 bowl
+rigged by the accuser; the 假孕 exposed by a prop) for a fresh payoff at the same budget.
+
+## 6. Dialogue
+
+A threat phrased as concern ("妹妹身子弱，可要当心"); rank in the pronouns (本宫, 臣妾, 嫔妾, 奴婢);
+punishment that sounds like mercy; truth said in public, meaning in private. Semi-classical but
+readable in a minute; the reversal line is a crisp trap-question ("那这香囊，为何在你宫里的绣娘手里？").
+
+## 7. What fails, and regulation
+
+降智 schemers (找事逻辑); evidence from nowhere; the emperor as deus ex machina; too many consorts too
+early; a rank that never moves; a heroine who only defends; an unforeshadowed betrayal. **Regulation**:
+宫斗 left prime-time satellite TV in 2011 and was targeted again in 2019; NRTA opposes 历史虚无主义; the
+2026 micro-drama measures ban distorting national history and require an AI label per episode;
+critics call 宫斗 anti-intellectual. **In practice**: a fictional dynasty, no real emperors as
+villains, scheming shown as costly, no cruelty to servants for gain, abortion, torture and 房中术 off
+screen, 巫蛊 exposed as a frame-up, never real magic (`story-advice` flags them).
+
+## 8. Sources (read as search summaries)
+
+知乎 on 宫斗剧法则, 女频宫斗 and 滴血验亲; CSDN on 短剧 structure; 界面 / 澎湃 / 36氪 on 宅斗; 豆瓣 后宫指南; 维舟 on
+宫斗 and 道德资本; 腾讯 / 搜狐 on 梅长苏's moves; 新华网 on 宫斗 and personality; 澎湃 / 虎嗅 (王倦); 腾讯 and 光明网
+on 《冒姓琅琊》; 腾讯 红榜 (《我在皇宫搞内卷》); 网易 on 《谁在说朕坏话》; 流媒体网 三点论 and 卡点; 电影界 on 红果
+编剧第一课; 360doc 编剧红宝书; 知乎 黄金手册; 三联 on 《甄嬛传》; 知乎 on 《女相师》; 百度百科 (《枕边刃》); CDT / 虎嗅 on
+15 years of bans; 澎湃 (2019); NRTA 2025-02 notice and 2026-06 draft measures; 澎湃 on 宫斗 and workplace; 腾讯
+/ 搜狐 on 降智; 中新网 (流潋紫); Strange Horizons on 琅琊榜; 界面 on 延禧攻略.

package/skill/short-drama/reference/story-revenge.md

@@ -0,0 +1,124 @@
+# Story genre `revenge` — 复仇 / 逆袭
+
+战神 · 赘婿 · 重生复仇 · 真假千金 · 扮猪吃虎 · 大女主复仇. The best-selling theme in 短剧 (DataEye
+2025: 逆袭 leads by heat and by title count; 打脸虐渣 in most top titles) and overseas (revenge,
+hidden identity, heiress on ReelShort / DramaBox). Shared machinery: `story.md`.
+
+## 1. The promise — certainty, not surprise
+
+The audience does not pay to be surprised; it pays for **a payoff it knows is coming**. The viewer
+sits in the god seat (信息差): they know the hero's real strength, watch the antagonist dig their
+own grave, and wait for "a certain result in a certain way". That is why 扮猪吃虎 is pleasurable
+even in the pig phase, and why every humiliation is a promise the drama must repay.
+
+- **先抑后扬.** Suppression buys the release; how much depends on the audience. Male-audience
+  shows want a fast rise with light humiliation; female-audience shows tolerate heavier
+  suppression because it deepens the reversal. Either way, never suppress without the audience
+  knowing a payback is owed.
+- **The curve is 憋屈 → 小爽 → 更憋屈 → 大爽**, and it is sliced into every episode, not spread
+  over the season.
+- **Regulators want it 爽而有度** (NRTA 2025): no venting-style 怼 / 打脸 for its own sake, no
+  全能正派 / 极端反派, no 以暴制暴 or 极端复仇, 重生 that follows real-world logic. Land the big
+  paybacks through exposure, law, wits and consequence, not private violence (`story-advice`
+  flags 私刑 / 以暴制暴 / 灭门 / 虐杀 in the outline).
+
+## 2. The series — a payback ladder
+
+| phase | where | what happens |
+| --- | --- | --- |
+| golden | E1 → the first paywall (卡一, E8–12 in an 80-episode run) | maximum humiliation **and one full 压→爽 loop by E03** against a tier-1 petty oppressor; the first identity hint; the paywall lands on a 身份 / 反转 / 危机 hook |
+| escalation | to ~75–80 % | cycles of 5–10 episodes: a higher antagonist presses harder (更憋屈), small paybacks accumulate, the cycle ends on a **big** payback and the next identity layer; the antagonists win some rounds (a setback near the midpoint and an all-is-lost at 70–85 %) |
+| climax | last 15–20 % | the arch-nemesis and the full reveal; the most public, most witnessed 打脸; closure through 情感释放 |
+
+**The antagonist tiers climb like game bosses (换地图):**
+
+| tier | who | leverage | falls |
+| --- | --- | --- | --- |
+| 1 | petty oppressors: the snobbish relative, the mother-in-law, the ex, the gatekeeper | contempt, gossip | early, often; the first small paybacks |
+| 2 | power holders: the rival heir, the fake daughter, the boss | money, position, family | across escalation, in big paybacks |
+| 3 | the arch-nemesis behind the original wrong | everything; hidden or untouchable | once, in the climax, at 80 % or later |
+
+**Tracked in the outline** (gate `story-revenge`):
+- `characters[].antagonist: { tier, firstPressEp, motive }` for every antagonist. `motive` is
+  required: an antagonist with no reason of their own is an 极端反派 and a 降智 villain.
+- `outline.paybacks[]: { ep, target, scale: small / big / final, by, how }` — every payback.
+  Every antagonist is finished (`final`) exactly once.
+
+**The rules** (researched defaults; waivable on the user's request):
+
+| rule | default | why |
+| --- | --- | --- |
+| `first-payback` | a payback 爽点 (打脸 / 复仇 / 身份揭露 / 智商碾压) in E1–3 | the golden three episodes close one loop |
+| `payback-gap` | ≤ 3 episodes in a row without one (≤ 5 when mixed with another genre) | 每集小爽点、三集大爽点 |
+| `setup-before-payback` | an antagonist's first payback comes ≥ 1 episode after they first press (≥ 3 for tier 2–3) | the payback is as strong as the pressure before it |
+| `tiers-ascending` | antagonists are finished in tier order; the top tier at ≥ 80 % of the run | the ladder climbs |
+| `big-payback-cadence` | in escalation, a big or final payback at least every 10 episodes (runs ≥ 20) | small reversals every 5–10 episodes, big ones every 20–30 |
+| `hero-delivers` | the arch-nemesis is finished by a lead | the backer (靠山) confirms the hero's power, never replaces it |
+
+## 3. The episode
+
+- **Hook in the first 3–5 s** (a humiliation, a threat, a secret shown); conflict by ~6 s.
+- **The 爽点 mid- or late-episode, then the cliff reopens the threat**: right after a payback,
+  a bigger problem, so the viewer does not leave satisfied.
+- **Episode 1** shows who the hero is, their status and their state through events, and puts a
+  concrete crisis on them (thrown out, stripped, threatened).
+- **A premise that sells has "a secret, a clock and a witness"**: something to reveal,
+  something running out, someone to walk in.
+- **The 打脸 beat**: the antagonist mocks (setup) → the reveal (the card, the call, the document) →
+  **the crowd reacts** (一定要有围观群众的反应) → new characters bow. On the storyboard this is the
+  `faceslap` grammar (`grammar-faceslap.md`).
+- **Revenge cliffs** by hook type: 身份 (someone kneels, a title is spoken), 危机 (the
+  antagonist's counter-move, a threat to family), 结果 (will they believe it?), 反转 (an ally is the
+  villain's mole), 道具 (the token, the medal, the call).
+
+## 4. Characters
+
+- **Protagonists**: the humiliated 赘婿 with a hidden 战神 / 神豪 identity; the reborn victim
+  with foreknowledge; the 真千金 who schemes (蓄势、谋势、借势) rather than wails; the 大女主
+  avenger who does the 打脸 herself. Give the hidden strength a *reason* to stay hidden.
+- **The backer** (a loyal subordinate, an old comrade, a kneeling general) is the witness who
+  walks in and confirms the hero's rank. In 大女主 dramas the man helps; she lands the blow.
+- **Antagonists stay credible**: real leverage (money, law, family, information), wins of their
+  own after the paywall, and disbelief that comes from pride or interest — never the 智障
+  villain who refuses evidence in front of him.
+- **Cast**: about 6–12 named, at least two antagonist tiers, one ally or backer.
+
+## 5. Keeping the stock beats fresh
+
+Stock beats: the humiliating banquet or wedding; the divorce-agreement slap; the phone call that
+makes a tycoon kneel; the black card / medal / token; the relative begging forgiveness; the fake
+daughter's frame-up; the reborn heroine pre-empting a trap; evidence played on screen before
+everyone.
+
+- **Rotate how the payback lands**: identity, wits (智商碾压), proof, a turned ally, the villain's
+  own trap backfiring. **Rotate who witnesses it**: family, the company, the whole society, the
+  press. **Let some paybacks cost the hero something.**
+- **Stack hot genres** (战神 + 赘婿 + 虐恋) and **play against type** (the obedient-looking
+  heroine who plots through information control). The market is moving from repeated hooks to
+  complete plots and distinct characters; the fixed "ch1 conflict, ch2 金手指, ch3 hook" opening
+  is now fatigue.
+- **Layer the identity reveal** (`story-advice`): at least three 身份揭露 episodes for a
+  hidden-identity premise, at most one in the golden phase, the full reveal in the climax.
+
+## 6. Dialogue
+
+Plain spoken language; 用最少的字让人物打脸. Lines of about 15 characters, no monologues. Many watch
+without sound, so the reveal must read as a picture (the kneel, the slap of the document). **Let
+the villain's taunt set up the exact words the payback throws back.** The hero's lines are terse
+and cold; emotion is saved for the 情感释放 beats.
+
+## 7. What gets rejected
+
+A weak first ten episodes (they decide approval); unoriginal characters; logic holes and
+dumbed-down villains; the formula opening; all the 爽点 spent early; literary lines; unshootable
+scale; venting-style or extra-legal revenge (regulatory).
+
+## 8. Sources (read as search summaries)
+
+流媒体网 卡点 (卡一 at E8–12) and 三点论; 电影界 on paywalls and 《无双》; hailuoshe on 憋屈—小爽—
+更憋屈—大爽 and reversal cadence; 知乎 短剧编剧黄金手册; 36氪/澎湃 on 爽 as the first principle;
+腾讯 on gradual 战神 reveals and 大女主 revenge; 投中网 on kneeling and the 智障 villain; 虎嗅 on
+episode-1 requirements; 澎湃 on 重生复仇; douban on the god seat; 界面 on male vs female 爽点; DataEye
+2025; NRTA 管理提示 "爽而有度" (2025); ScreenWeaver (hook/turn/cliff, four tent-poles, secret-clock-
+witness); InkTip; John August on microdramas; Save the Cat (Fool Triumphant, Superhero); revenge
+tragedy conventions; thewritepractice on revenge and underdog stories.

package/skill/short-drama/reference/story-scifi.md

@@ -0,0 +1,112 @@
+# Story genre `scifi` — 科幻
+
+末世 / 末世重生囤货 · 时间循环 · 穿越时空 · AI 觉醒 · 克隆 · 赛博朋克 · 星际. This is the **story** side; the
+look is `scifi-hard` / `scifi-soft` (`cinematography-scifi-*.md`). Shared machinery: `story.md`.
+
+## 1. The promise — one "what if", followed all the way
+
+- **The novum as a contract**: change one element and work out what follows — the 核心设定 and its
+  衍生设定 across technology, society and people (Suvin's novum, "cognitive estrangement": the
+  impossible reported as plain fact).
+- **Four emotions sell**: wonder (奇观 — and AI makes a rocket cost the same as a cup), dread of the
+  future (Black Mirror: technology amplifies a human flaw — "a human problem, not a technological
+  one"), the 爽 of foreknowledge (末世重生囤货: reborn before the disaster, sell the house, build the
+  fortress, stock up, the 空间 cheat, refuse the relatives who betrayed you), and mastery through
+  repetition (the time loop as a game level replayed until beaten — 《开端》).
+- **The market**: 男频 live action fell a third from its 2024 peak while **末世 漫剧 topped the Douyin
+  charts**; 三幻 (sci-fi, fantasy, supernatural) lead AI drama because effects are cheap; overseas
+  apps run apocalypse-vault and time-travel lines.
+
+## 2. The series
+
+- **Wells' law**: one impossible assumption, developed with the strictest logic — "a rigid exclusion
+  of other marvels". **Sanderson's first law**: resolving a conflict with a power satisfies in
+  proportion to how well the audience understands it; **the second**: limits and costs are more
+  interesting than powers.
+- **The arc**: the novum (or its consequence) in E1 — the 末世 pattern opens on a past-life death or
+  betrayal, the rebirth, and **an explicit countdown** ("the cold wave in 30 days"); two or three rules
+  through action → **escalation** of scope (personal → building → city → world; heat → cold → quake →
+  fog), each tier adding or testing one rule → **the midpoint turn** (the countdown hits zero, or a rule
+  proves incomplete — the loop has a second trigger, the AI a hidden directive, someone else was
+  reborn too; in 《开端》 sleeping triggers the loop as well as dying) → **the climax** resolved through a
+  rule the audience knows; a loop breaks because the character changed (Edge of Tomorrow breaks by
+  sacrifice).
+- **Hard vs soft**: 硬科幻 makes the idea the subject; 软科幻 uses it for plot and emotion. 短剧 is almost
+  always soft sci-fi with one hard-feeling rule.
+
+**Tracked in the outline** (gate `story-scifi`): `outline.brief.novum: { statement, category (末世 /
+时间循环 / 穿越 / AI / 赛博 / 星际 / 克隆 / 其他), mode, firstEp }`, `outline.worldRules[]: { id, rule, kind:
+power / limit / cost / trigger / exception, shownEp, decidesEps[], brokenEp?, revealSetupEp? }`,
+`outline.countdown: { label, zeroEp, log: [{ ep, value }] }`, `outline.loops[]: { n, ep, trigger,
+newInfo, changed }`, `outline.foreknowledge[]: { event, plantedEp, payoffEp, deviated }`,
+`outline.secondaryNovums[]: { id, derivesFrom }`.
+
+| rule | default | why |
+| --- | --- | --- |
+| `novum-first` | the novum stated, on screen in E1 | the genre promise lands in the hook |
+| `one-novum` | at most one marvel that does not derive from the novum (rebirth + system is one package) | Wells' law |
+| `rule-costs` | at least one limit or cost rule per power rule | Sanderson's second law |
+| `show-before-use` | a rule is shown before it decides a turn; a climax-deciding rule ≥ 3 episodes before the climax; a rule bends only as a set-up exception | Sanderson's first law; no contradictions |
+| `countdown` | a 末世 drama runs on one; it only counts down; on screen in ≥ 60 % of the episodes before the climax | the spine and the free cliffhanger |
+| `loops` | every loop brings new information, skill or cost; never two in a row where the hero changes nothing | repetition without new context is monotony |
+| `foreknowledge` | paid off within 10 episodes of being planted | the 爽 of knowing |
+| `midpoint-turn` | at 40–60 %: the countdown reaches zero or an exception / trigger rule appears, on a 反转 / 危机 hook | the rule proves incomplete |
+
+## 3. The episode
+
+- **Hooks from the novum**: 危机 (the countdown, the disaster through a window), 道具 (the system
+  panel, the 空间 glow, a device), 结果 (a future outcome — "in the last life she died here"), 反转 (a
+  rule exception).
+- **The countdown on screen** (a HUD number) gives a free cliff every episode and cheap continuity.
+  State the rules early and repeat them.
+- **Loop episodes**: show the reset fast (a 3–5 s repeated motif), spend the rest on what is
+  different.
+- **One piece of foreknowledge pays off per episode** ("last life X happened, so this time…") — it
+  fills the 智商碾压 and 打脸 slots. Add a butterfly-effect deviation by the midpoint so foreknowledge
+  can go wrong.
+
+## 4. Characters
+
+- **The ordinary person facing the novum** carries it (郭帆's tests: does it relate to "me"? a strong
+  everyday metaphor? a character with emotion?). The Chinese sci-fi emotional core is family.
+- **The 三体 lesson**: keep one idea intact and simplify everything else.
+- **The expert** shows a rule working, paired with a skeptic so explanation comes out as argument.
+- **The AI character**: one explicit directive and one limit; its awakening a choice that breaks or
+  reinterprets the directive, set up earlier.
+- **末世 antagonists**: the past-life betrayer, the neighbours who 道德绑架, the scavengers — the novum
+  makes human greed visible.
+
+## 5. Premises and freshness
+
+末世重生囤货 (fresh now through an odd angle — running an apocalypse business with zombie employees,
+raising a child, a cozy or comic tone); 时间循环 (fixed-start, death-triggered, self-consistent);
+穿越 (mostly modern skills in the past — respect history); AI 觉醒 / 克隆 / 意识上传 (identity drama that pairs
+with 身份揭露 — ration "it was all a simulation" to once); 外星入侵 / 星际 (one location per arc);
+规则怪谈 (a model for the rules ledger).
+
+## 6. Exposition
+
+Need-to-know, when it matters to the point-of-view character; images carry it; no "as you know".
+**Rule through consequence**: someone tests the rule, a visible cost, then one line names it —
+from a panicking character, not a narrator; or the rule on screen as text (a panel, a warning sign, a
+HUD). **One new rule a scene**, one or two an episode. Don't explain *why* the loop happens if its
+rules are clear (《开端》 never does). `story-advice` flags lecture lines (众所周知 / 你知道的 / 原理是).
+
+## 7. What fails, and regulation
+
+A climax solved by a rule never shown; stacked marvels; a contradicted rule; spectacle without a
+personal stake; loops without new information; foreknowledge that never goes wrong; "science" that
+dates; copy-paste 末世. **Regulation**: the 2026 micro-drama measures ban rendering violence and terror
+(zombie gore, scavenger violence), false information (no invented "facts" about real diseases or
+disasters), superstition (keep the novum science), harm to minors, and historical nihilism (no real
+countries as invaders, no mocking real historical figures in 穿越). The 2025 "微短剧+" plan encourages
+科普 dramas; the 2020 科幻十条 signals goodwill. Keep 末世 hopeful and cooperative by the end.
+
+## 8. Sources (read as search summaries)
+
+SFE on Wells' law; Strange Horizons on Suvin; Brandon Sanderson's laws; 中国作家网 on 设定 and 硬/软科幻, and on
+《开端》; 澎湃 on 《开端》; Overthinking It and Dazed on Black Mirror; Uncanny and Arc Studio on exposition;
+time-loop craft (Prose and Quans, PremiumBeat, 百度 TA说); 界面 (郭帆) and 观察者 (《流浪地球3》); 中新网 and 界面
+on 三体 adaptations; 豆瓣, 腾讯 and 网易 on 末世囤货; 电影界 on the 抖音 漫剧 chart; 第一财经 and TVtalk on AI drama;
+澎湃 on 红果 2025; 百度百科 (《觉醒》); 维基 on 规则怪谈; DramaBox; 国家电影局 and 界面 on 科幻十条; 新华网 and 南方网 on
+the 2026 measures; 新华网 on tiered review; 澎湃 on 微短剧+; 中新网 on the 2011 time-travel restrictions.

package/skill/short-drama/reference/story-sweet.md

@@ -0,0 +1,128 @@
+# Story genre `sweet` — 甜宠
+
+霸总甜宠 · 先婚后爱 · 契约婚姻 · 闪婚 · 豪门甜宠 · 古装甜宠 (追妻火葬场 is a neighbour with its own
+ratios, §7). Female-audience content is over 70 % of 短剧 heat, 现言甜宠 is a top-3 genre on
+Douyin, and about three quarters of ReelShort's audience is female. Shared machinery: `story.md`.
+
+## 1. The promise — to be seen and chosen
+
+The fantasy is being **seen and chosen** by someone who could have anyone, and mattering more to
+them than their power: 被偏爱, 双标 (cold to the world, soft only to her). The genre signature is
+**高糖低虐 + 强反差**. Three layers of sugar, and a good series gives all three and escalates them
+from private to public:
+
+| layer | what it is |
+| --- | --- |
+| 糖 | the single sweet moment (the protective arm, the jacket, the kiss) |
+| 心动 | the tension before it (拉扯: 相爱不敢认，想近不敢亲，嘴硬心软，欲言又止) |
+| 被偏爱 | the lasting reward: public, exclusive, unconditional favour |
+
+The taste is moving from the domineering 霸总 to the **外冷内热 / devoted "puppy" male lead** — the
+superior party bowing down for love — and from the 小白花 to a heroine who thinks for herself.
+The power balance shifts over the series: he starts in control, they end as equals.
+
+## 2. The series — milestones on time
+
+The romance beat sheets agree on the shape (Gwen Hayes' four phases — set-up, falling in love,
+retreating, fighting for love; Billy Mernit's Cute Meet → Hook → Swivel → Dark Moment → Joyful
+Defeat; the K-drama confession at the midpoint). Compressed to 短剧 paywall realities (80 episodes):
+
+| phase | episodes | milestones (`outline.romance[]`) |
+| --- | --- | --- |
+| golden | 1–10 | **meet** in E1 and the **bind** by E3 (闪婚 / contract / 替嫁 / cohabitation / 婚约 — the device that keeps them together); the first **spark** and **touch** by E5; the first **protect** (护妻); the paywall (卡一 ~E8–12) on an emotional or identity hook |
+| escalation A | 11–40 | forced proximity; **jealousy**; the first **kiss** around 25–35 %; he defends her in public; the rival escalates |
+| midpoint | ~40–60 % | the **confession** — the "midpoint of love", a relationship acknowledged or a 隐婚 half-revealed |
+| escalation B | to ~70 % | the family obstacle, 掉马 (an identity reveal), doubt; short **misunderstandings** that clear within 2–3 episodes |
+| climax | last 25 % | **one** short **black moment** at 70–90 %, then **reconcile** within 5 episodes, the grand gesture, **public** (公开 / 官宣 / wedding), the obstacles 打脸-ed, an epilogue of sugar — **commit** |
+
+Milestone vocabulary: `meet` `bind` `spark` `touch` `jealousy` `protect` `kiss` `confession`
+`misunderstanding` (with `until`, the episode it clears) `reconcile` `public` `black-moment`
+`commit`.
+
+**The rules** (gate `story-sweet`; researched defaults, waivable on the user's request):
+
+| rule | default | why |
+| --- | --- | --- |
+| `sweet-share` | ≥ 40 % of episodes pay in sugar (强宠 / 情感释放); ≥ 25 % when mixed | 高糖低虐; producers literally count kisses |
+| `sweet-gap` | ≤ 3 episodes in a row without sugar (4 in the climax; +2 when mixed) | 误会 cycles close in 2–3 episodes; "虐后快速给甜" |
+| `meet-bind` | meet in E1, bind by E3 | episode 1 ties the leads together; without the bind they drift apart |
+| `first-touch` | touch (or kiss) by E5 (runs ≥ 10) | show the promise before the viewer pays |
+| `first-kiss` | by half the run (runs ≥ 10) | Hayes' deepening-desire phase; the K-drama midpoint |
+| `midpoint-confession` | a confession at 40–60 % (runs ≥ 20) | the midpoint of love |
+| `misunderstanding-run` | each clears within 3 episodes (5 in the climax); only one may run past 3 | "a misunderstanding one sentence could clear" loses the audience |
+| `black-moment` | at most one, at 70–90 %, mended within 5 episodes | Break Up → Grand Gesture; Dark Moment → Joyful Defeat |
+| `he-ending` | sugar in one of the last three episodes | the genre promises a happy ending |
+
+`story-advice` also notes a climax without a **public** milestone and 强宠 payoffs that are mostly
+money (黑卡 / 买下 / 豪车 — regulators target 拜金 / 炫富 / 炫权 爽点).
+
+## 3. The episode
+
+- **Hook** in the first seconds, readable with the sound off: tension or friction between the
+  leads, or a rival's pressure.
+- **The sugar lands in the back half, just before the cliff** (the 7-3-1 model: setup, high sugar,
+  a sting of suspense). Sweet shots are held longer (3–6 s, close) than 打脸 shots.
+- **A skeleton**: 0–10 % hook → 10–60 % pressure (rival, family, situation) → 60–90 % 撒糖 or
+  护妻 → 90–100 % a new question.
+- **Cliffs**: 情感 (the almost-kiss interrupted; "你吃醋了？" unanswered; he overhears her plan to
+  leave), 身份 (his identity or the 隐婚 about to be exposed), 危机 (the rival's or the family's
+  ambush), 反转 (the contract was his long-planned choice — 蓄谋已久). Romance cliffs work best when
+  desire and status collide in public.
+- **Plan one or two shareable 名场面** per series (the 抵墙吻) — they drive 二创.
+
+## 4. Characters
+
+- **The male lead** is a *position*, not a job title — anyone 上位者 (CEO, general, prince,
+  lawyer, 大佬). His harshness is aimed at others, never humiliating her; his possessive acts
+  protect, never control. Overseas audiences reject possessiveness framed as love, non-consent and
+  humiliation.
+- **The female lead** is independent and has wins of her own (she lands some 打脸 / 智商碾压 — not
+  only rescued). Avoid the greasy heroine (exaggerated coquetry, over-explicit flirting).
+- **The map**: the rival 女二 (schemer or 白月光, paid off by 打脸); the gentle 男二 who must not
+  out-care the lead (second-lead syndrome — give him a noble exit); the family obstacle (豪门
+  mother, stepmother, elders); an ally or comic relief (assistant, best friend, 萌宝). Keep the
+  core to 4–6.
+- **Avoid**: a 一夜情 opening; money as the sweet payoff; the male lead as the main antagonist
+  (only in 追妻火葬场).
+
+## 5. The tropes, kept fresh
+
+Staples: 壁咚 / 抵墙吻, 公主抱, the piggyback, 包扎伤口, 披外套, 吃醋, 护妻 (撑腰 in public), 双标,
+forced proximity (同住 / 一张床), 隐婚 → 公开, 掉马, the contract deadline, the family-dinner defence.
+
+- **缺啥给啥** — small, specific, personal details beat grand gestures.
+- **Borrowed intimacy** — the fake couple's staged touches become dangerous because they are real;
+  a contract is a slow burn with a deadline: use the clock.
+- **Stack a primary and a secondary trope** (contract marriage + hidden identity); mix genres
+  (萌宝, 年代, 古装, 奇幻).
+- **Run each trope once privately and later publicly, and flip who does it** (she protects him,
+  she makes him jealous). Never the same trope in consecutive sweet episodes.
+
+## 6. Dialogue
+
+- **拉扯**: cold words, soft details (口是心非); the truth in micro-expressions, eye contact and small
+  actions. Care through action and one-line understatement ("手给我"), not speeches.
+- **句句不骚，句句都撩.** "关系没到位，再高级的情话也会变成油腻" — match the line to the relationship
+  stage: no "你是我的" before the midpoint. Keep 土味情话 for comic self-awareness.
+- **Banter carries status play**: she wins verbal rounds.
+
+## 7. What gets rejected — and 追妻火葬场
+
+Rejections: old patterns (90 % of scripts cut at first review for 模式老套); everyone stupid
+(全员降智); no 拉扯, flat chemistry ("chemistry matters more than plot"); misunderstandings a line
+would clear; too long 虐; a toxic, controlling or violent male lead; 炫富 / 拜金 payoffs; greasy
+flirting; 工业糖水 sameness.
+
+**追妻火葬场** (the chase after he hurt her) runs about 30 % 虐, 40 % his regret, 30 % the chase —
+his regret must outweigh her suffering; she is soft early and strong later. Waive
+`misunderstanding-run` / `sweet-gap` for it explicitly (`storyExceptions`), on the user's word.
+
+## 8. Sources (read as search summaries)
+
+电影界 / 万众编剧 on 甜宠 (高糖低虐, 7-3-1); 知乎 短剧编剧黄金手册 (误会 in 2–3 episodes, sweet shot
+lengths); 36氪 on 霸总 as a position and 大女主; 澎湃 / 腾讯 "从霸总到训狗"; NRTA 霸总 notice (2024);
+流媒体网 / 投中网 / CSDN on 卡点 (the 甜宠 paywall at E5 is a single weak source); 知乎 on CP 拉扯, 追妻
+火葬场 ratios, 甜宠 苏点; 澎湃 拉片 《咬清梨》 and "女主越来越油"; 虎嗅 on script rejection; DataEye
+2025; 21经济网 on CP stickiness and counting kisses; 三联 on padding; Gwen Hayes (Romancing the Beat);
+Billy Mernit; K-drama conventions (TV Tropes, Dramabeans); Filmustage; Rolling Stone on ReelShort;
+ScreenWeaver; the Vertical Drama Love / Deadline fan survey; fake-relationship craft.