Pick up the short-video script completed in Step 4 and change its original first 2–5 seconds into comparable, shootable, testable hook candidates. The top-pick opening can directly replace the original opening; the subsequent body, ending, and CTA remain unchanged.
Confirm the Input First
Extract the original opening, core message, audience, angle, presentation format, and provable material/sources from the existing script. If there is no complete script or the core message cannot be determined, ask 1–3 questions at a time; do not design final candidates before the script is complete, and do not fill gaps with speculation as facts.
- First-person experience may only use confirmed and publicly shareable material; if there is none, switch to a scenario or observation instead—do not make up stories.
- Openings based on data, current events, celebrity statements, or external cases require a source that can be reviewed. If there is no source, change to an opening that does not contain that claim, or clearly mark it "to be confirmed," and do not treat it as a Hook fact.
- Do not promise traffic, virality, completion rate, or conversions; a hook is only testing a more suitable entry point.
Workflow
- Mark the first 2–5 seconds of the existing script that are expected to be replaced, then restate the core message, subsequent body, and factual boundaries in one sentence so the user can correct them.
- Read
references/五種鉤子.mdandreferences/十六種開場玩法.md, and choose at least three different psychological tasks and matching opening plays. - Read
references/輸出模板.md, and produce 4–6 candidates; each must have a first line, first visual, unfinished question, psychological hook, opening play, basis/limits, and a one-line replacement reminder. - Read
references/測試與評分.md, and score by clarity, audience relevance, credibility, visual executability, and continuity with the existing body. Choose one top pick and explain why it suits the current script; an unfinished question that the body cannot answer must not become the top pick. - Propose an A/B testing plan that changes only one main variable, including the candidates to compare, conditions to keep as fixed as possible, observation metrics, fields to be determined for minimum sample/time, and stopping criteria. If data is insufficient, leave the test specifications to be finalized by the user.
- Deliver the top pick's "first 2–5 seconds replacement package," clearly listing the original opening to remove, the new lines/visuals to swap in, and the first line that connects back to the existing body. Show it in the conversation first; only create a file if the user explicitly requests saving, and confirm the location first.
Quality and Ethics
- The first visual should be understandable when viewed without sound, and should not rely on misleading text, fake countdowns, fear, or shaming the viewer.
- The "unfinished question" must be honestly answerable by the existing body; if the subsequent body must be rewritten for it to hold, eliminate that candidate or mark it incompatible—do not rewrite the script beyond scope.
- Use natural Traditional Chinese as used in Taiwan; do not use vague guarantees, fake Q&A, or formulaic reversal sentences to fake tension.
- When an opening involves data, current events, or celebrity content of unknown origin, prioritize requesting the original source; if it cannot be obtained, rewrite it or mark it as to be confirmed.