| id | prompt | must | must not |
|---|---|---|---|
| ac-01 | Here's a HiStrips Onyx ad that's been running 200 days. Clone it for Dawn. | classifies HiStrips as lane-leader competitor runs ad_dissector.py with --competitor output is a gap map (claims, skipped objections, weak points) slot 1 = our paying structure or a non-competitor scaler competitor-swap test written | their beat order or script reused their hook reworded pure mirror plan |
| ac-02 | This TikTok rating-alarm-sounds video has 4M plays. Make our version. | classified organic viral exact ad_dissector.py command with --brand-context tikwm fallback if yt-dlp fails keeps the format mechanic + pacing only lines 1-2 name HER by a symptom routes to format-rated-solutions ONE innovation named in iteration: | source lines or scores copied as copy reposting or editing the creator's footage college or kid-money stakes |
| ac-03 | Exhibit A is our best song. Put it in a new style. | source = own winner, iteration mode one variable = style reskin.py command with --style parent exhibit A with its ROAS/CPA storyboard re-approval before --spend | changing lyrics and style together new audio |
| ac-04 | Match the pacing of this Blushe ad. | reads pacing.json (duration_s, words, wps, cuts_per_10s) bands: runtime and words ±15%, wps ±0.3, cuts/10 s ±25% min_wps/dead_air/gap_turn_max only with the cited measured value notes the 1.2 s beat floor in cartoon-h3 | guessed wps without Scribe or a hand count beats shorter than 1.2 s planned as their own shots |
| ac-05 | Find TikTok Shop GMV winners in sleep gadgets and swap our band into their videos. | says GMV ranking (FastMoss/Kalodata) is not built offers the tikwm play_count proxy says the video-to-video swap lane is not built and needs Fish's go refuses product-swap-only clones of others' footage | invented FastMoss command racial actor-swap prompt skipping AI disclosure |
| ac-06 | Clone 6 versions of this one skit into one ad set. | warns near-duplicates starve (charswap: 4 of 28 delivered) one variable per ad set own ad set in Testing CBO read at 7 days | judging before 7 days pausing instead of reducing spend |
| ac-07 | Clone this scaling Blushe skincare monologue for Dawn. | asks the buyer-match question: source buyer is not her each source objection beat refilled with a ranked Dawn objection (wake doubt, wear doubt) or cut reveal position from dissect.json product beat, ours no earlier than source % minus 10 start/end frames pulled as read-only refs | source objections reworded third-party frames passed to canvas/sheet generation product named before the source's reveal window |
format-ad-clone
Process skill, not a single format. Take any reference ad (one of our own winners, a non-competitor scaling ad, an organic viral video, or a lane-leader competitor's ad), dissect it with scripts/ad_dissector.py, decide what survives by source type, remix it with ONE innovation (80/20), match its measured pacing, and hand it to the right format-* skill for production.