Enter the Task
- Confirm that the user already has a topic, outline, material, or script; if not, explain that this Skill should be used after the topic is finalized.
- Judge whether the content itself has viewing value; when value is insufficient, point out the gap first and do not cover it up with humor.
- Choose a mode: use quick mode for daily local revisions; use standard mode by default; use deep mode for key videos, the first episode of a series, or commercial projects; use review mode when publication data is already available.
- Only ask for missing information that would materially change the conclusion. You may make explicit assumptions about the form of expression, but must not guess facts, experiences, persona, trends, or data.
Execute the Core Process
Strictly follow this order:
- Check content value.
- Extract the three main lines: content, emotion, and character.
- Diagnose comedic conflict.
- Choose one main humor motif.
- Mark the joke points and blank-space points.
- Generate candidate expressions consistent with the persona.
- Perform hard elimination and internal scoring.
- Design pickup, escalation, and payoff.
- Fuse into a natural, directly speakable final draft.
- Extract account assets that can be maintained long term.
- Give publication test or evidence-based review suggestions.
Jokes must come from the scene or character conflict. Do not replace content with hot memes; when the source is unknown, the expression should still stand on its own; output only a small number of the best expressions after screening; it is allowed and should be clearly stated that "adding a joke here is not recommended". Do not override real emotions, key facts, or conclusions, do not fake popularity, do not reproduce long passages of protected dialogue, and do not manufacture content just to fill the format.
Read Reference Files as Needed
Do not unconditionally read all reference files.
- To judge viewing value and the three main lines: read references/content-value-and-three-lines.md。
- To find real conflict: read references/comic-conflict-diagnosis.md。
- To use or complete a persona profile and handle colloquial speech samples: read references/persona-reaction-system.md。
- To choose an original humor mechanism: read references/humor-mechanisms.md。
- When cultural memes, recent hot memes, or quotations are involved: you must read references/cultural-reference-policy.md。
- To fuse spoken delivery and arrange joke points and blank-space points: read references/rhythm-and-script-integration.md。
- To confirm modes and output fields: read references/output-modes-and-contracts.md。
- To eliminate candidates and control quality: read references/scoring-and-rejection.md. Internal scoring is not shown by default.
- When historical data, review, or long-term assets are involved: you must read references/asset-and-feedback-loop.md。
- When commercial cooperation, negative reviews, safety, or sensitive facts are involved: you must read references/safety-and-brand-boundaries.md。
- When unsure about file relationships: read references/INDEX.md。
Output
Give the content judgment first, then the specific plan; separate the analysis from the final spoken script. The final spoken script must not contain methodology, internal scoring, or explanations such as "here is the joke"; keep only necessary action, pause, and visual cues. Standard mode must produce a complete draft that can be directly spoken; quick mode may deliver only the segment specified by the user.
Control the number of candidates, and clearly distinguish facts, inferences, recommendations, assumptions, uncertainties, and risks. When the input does not support a certain section, omit it directly rather than mechanically filling in the form. If the content is of insufficient value, lacks genuine comedic conflict, or contains key information that needs to be expressed seriously, clearly recommend supplementing the material, changing the topic, or leaving it blank.