Subtitle Remix Pipeline

Purpose

Convert subtitle material into a complete, original, market-fit adapted script without losing source information or copying protected expression.

This skill combines two stages:

1. Subtitle calibration: preserve and correctly attribute the source.
2. Original adaptation: extract the hook, rebuild the story, and deliver a new complete script.

Do not skip either stage unless the user explicitly says the calibrated script already exists and is reliable.

Hard Gates

Do not deliver the final adapted script until all gates pass:

• Source file inventory is recorded.
• Clean subtitle source is selected as authority when multiple files exist.
• Original episode count is verified or reasonably inferred.
• Missing subtitle line count is 0 for the calibration stage.
• Speaker attribution audit passes.
• Cross-story/trailer residue is excluded or clearly marked.
• Adapted work is substantially original, not a renamed or lightly rewritten copy.
• Full adaptation means full episode count unless the user explicitly asks for a sample or compression.
• Script density must roughly match the source. If the source averages about 40-50 subtitle lines per episode, each adapted episode must contain comparable dramatic content: multiple action beats, dialogue turns, reactions, transitions, and a hook. A few-sentence episode summary is an outline, not a completed script.
• Final self-score is at least 90/100.
• Long final output is written to a local .md file and imported to Feishu/Lark by default.

Automatic fail:

• Outputting only 3-5 sample episodes when the user asked for the full adaptation.
• Arbitrarily reducing a 70+ episode source to a small number of episodes.
• Outputting full episode count but making each episode only a few lines long when the source episodes contain dozens of subtitle lines.
• Speaker labels are visibly wrong, including self-address errors.
• A character or name appears that was not in the source or the new adaptation plan.
• The protagonist's money-making path does not match the chosen era/industry.
• A requested system/golden-finger appears at the wrong trigger point or replaces protagonist skill.
• The score is written as a decoration but low-scoring problems are not fixed.

Workflow

1. Source Inventory

For every source file, record:

• Absolute path.
• File type and encoding.
• Rough/clean status.
• Character count with whitespace.
• Character count without whitespace.
• Non-empty line count.
• Inferred episode count.
• First and last meaningful line.
• Suspected contamination, trailer residue, or unrelated character names.

If the user provides both a rough file and a clean file, use the clean file as the authority, but remember what was wrong in the rough file.

Never silently merge unrelated tail content into the main story.

2. Subtitle Calibration

Create a strict source script before original adaptation.

Requirements:

• Preserve every meaningful subtitle line.
• Convert each line into dialogue, OS, narration, screen text, action, reaction, or transition.
• Add playable scene description and action beats; do not output a bare transcript.
• Match the source episode count unless the user asks otherwise.
• Keep scene order aligned with the source.
• Use stable playable labels for unknown speakers.

Speaker audit rules:

• A name inside a line is often the addressee, not the speaker.
• Prevent self-address errors such as a character saying their own name without context.
• Restrict each scene to characters who are actually present.
• If the line says “爸爸/爷爷/卫东/小柔/圆圆” etc., infer who is being addressed before assigning speaker.
• If uncertain, use a stable group label and note the uncertainty instead of inventing a named character.

Coverage audit:

• Count all non-empty source lines.
• Confirm every line is represented.
• Missing line count must be 0.
• If a line is merged, it must still be traceable.

3. Hook And Risk Diagnosis

From the calibrated script, extract only abstract mechanisms:

• Audience hook.
• Emotional engine.
• Structural engine.
• Protagonist pressure.
• Antagonist pressure.
• Escalation rhythm.
• Payoff type.

Create a must-not-copy list:

• Original names.
• Exact relationship setup.
• Scene order.
• Signature set pieces.
• Unique props/rules.
• Distinctive dialogue.
• Iconic visuals.
• Ending mechanism.

Use copyright-safe abstraction. Do not preserve exact expression.

4. Market And User Constraint Calibration

Respect the user's latest direction over earlier drafts.

If the user asks for current platform fit, verify current platform/genre tendencies before selecting the adaptation direction. For Red Fruit/Hongguo-style manhua drama, identify a practical genre match instead of forcing a random trend.

For period-realism adaptations:

• Keep the chosen era's economy, transport, documents, food, business, and social relations plausible.
• Profit must come from a believable chain: skill, legal permission, processing, transport, buyer, contract/order, pricing pressure, risk.
• The male lead's distinctive skill must be visible in repeated actions.
• Generic scavenging or simple collecting cannot justify large dividends unless the business logic is rebuilt.

For system/golden-finger adaptations:

• Define when the system first appears.
• Define who can see it.
• Define what it reports.
• Define what it cannot solve.
• Keep protagonist skill necessary.
• If the user says the system appears only after a specific event, that trigger point is mandatory.

5. Original Rebuild

Rebuild at least these elements:

• Era/social environment.
• Industry/business domain.
• Core resource.
• Protagonist identity, wound, goal, and active skill.
• Antagonist power base.
• Character relationship web.
• System/rule/business mechanism.
• Main locations and visual symbols.
• Climax and public reckoning arena.

Changing names is not enough. The new world must create different choices, conflicts, scenes, and payoffs.

6. Full Script Draft

For a full adapted script, include:

1. 改编策略说明
2. 原作吸引力机制
3. 避险替换表
4. 原创世界观与人物小传
5. 原创故事梗概
6. 分集大纲
7. 剧本正文
8. 自查自纠报告与评分

For long sources, write the full deliverable to a .md file. Do not rely on chat-only output.

If the source has a known episode count, output the same adapted episode count unless the user explicitly requests compression or expansion.

Density rule:

• Estimate average source subtitle lines per episode.
• For final adapted script, target a similar number of playable script lines per episode.
• A normal adapted episode should include at least one scene heading, character list, 8-15 visual/action beats, 15-30 dialogue or OS/screen-text turns, and a clear episode hook.
• If time or context requires staged delivery, label the output as a batch draft and continue expanding; do not call it final until all episodes meet the density target.

7. Self-Review And Revision

Score using this 100-point gate:

• Source coverage and episode integrity: 15
• Speaker attribution and source integrity: 15
• Episode script density and completeness: 10
• Hook extraction and audience value: 10
• Original worldview and causality: 10
• Character goal, skill, and active choices: 15
• Conflict escalation and payoff: 10
• Setup/rule consistency: 10
• Copyright-risk distance: 10

If any hard gate fails, the score must be below 90. Revise before delivery.

Do not present a below-90 draft as final.

8. Feishu/Lark Delivery

After generating a long final script file, import it into Feishu/Lark as a docx unless the user explicitly says not to.

Use a relative file path from the file's directory:

lark-cli drive +import --as user --file ".\\final-script.md" --type docx --name "document-title"

Return:

• Local file path.
• Feishu/Lark URL.
• Final score.
• Key self-check results.

If import fails, explain the exact blocker and keep the local file available.

Final Response Checklist

Include:

• Authoritative source file used.
• Episode count.
• Subtitle coverage result.
• Speaker audit result.
• What hook was extracted.
• What was rebuilt.
• Why it is not merely a renamed copy.
• Final score.
• Feishu/Lark document URL for long deliverables.