Copy Engine

I had six scripts checking my copy. Clarity, cliches, spam words, CTA strength, compliance, coordination. A live long-form draft passed all six and still read like a memo, and nothing in the skill could tell me why. copy-engine drafts, edits, scores and audits copy for any surface, and it never writes from training data. Every draft is modelled on one of 43 real exemplars that carry their own source and evidence class, then gated by 8 scripts that print PASS or FAIL with a number. v0.6.0 added the register where the writer is the hero: the Epiphany Bridge script from Expert Secrets pp.213-222, the Hero's Two Journeys from pp.197-211, and verify-narrative-voice, which measures four bars. The draft that read cold scored 17 em-dashes, 16.8 first-person per 1k, 11.1 second-person per 1k and 13.4-word sentences. Four of four bars failed. The story rewrite of the same material scored 0, 39.8, 26.8 and 9.9, and passed all four. v0.7.0 writes video scripts as a timed beat table: the spoken words and the on-screen text for each beat, in each language. My own section-designer demo put 67.5 seconds of voiceover into a 60-second video, with 5 of 7 beats over their window. Rewritten words only, it passes video-director's clock in English and Russian, 0 of 7 over. Guidance-only, so there is no backend to go down. It reads voice.md from @di-atomic/voice-builder, hands titles and video hooks to hook-generator, hands the beat table to video-director's clock, and persists through the OPVS coordination peers.