Non-negotiables (every state, every host)

1. Copy the reference. Study the whole reference, picture and sound. Keep its people, action, camera, hook, length (±25%) and its shot and setup counts (±30%, from scripts/reference_shots.py); list every change in changes_from_reference, shown first at the script approval.
2. Match its sound and pace. Speech, music and effects follow the reference. Silence only when the reference is silent or the owner asks for it. Never a sine tone, a drone or an empty track. scripts/reference_rate.py measures the reference's words per second (never estimate it); a voice more than 10% slower is regenerated or sped up (at most 1.25×) before lip-sync and picture timing.
3. People and demos stay people and demos. Plan a person, hands or a live demonstration in the reference as one; never plan a text card, dots or a still in its place. One that ends up on screen anyway is an ask finding you recommend fixing.
4. The script comes from write-video-ad-script. Fetch it with catalog_fetch; never write the script inline.
5. Use the real product. Its real photos, its real size and its usage steps. A still packshot gets the real photo through overlay. A pack held or used on screen is made with openai/gpt-image-2/edit, the real product photo as reference image 1, front label facing the camera. A blank, unprinted or label-less brand pack is must_fix. Every use step and product state (dry contents, cooked dish, strip on the nose) comes from a real brand photo sent as a reference.
6. Keep the approved look. Camera, focus, lighting, texture, setting and finishing follow the brief/reference. Phone realism belongs to phone footage; studio, cinematic, stylized and animated formats keep their own look. Preserve product truth and plausible anatomy in every format. Use libraries, look and rubric.
7. Review the final against the reference, listened or measured, never claimed. scripts/qc_evidence.py must exit 0 on the exact final bytes (it fails only a broken file); then mark the video complete and read the host's final check.
8. Two owner approvals, both in this chat: the script and the ingredients. In review_mode: auto one yes to the full total covers both (see check severity). Agent checks add no other approval step.
9. Quality warns, never blocks. Every finding is must_fix (always redone, never delivered), ask (shown with the price of the fix; in review_mode: auto you decide) or fine (kept). Only money, consent, safety and broken files stop the run. A redo is paid from the approved total when it fits, otherwise after a new yes to its price; never drop a scene. See check severity.

Reading rule. Read one file per call. If the output says it was truncated, re-read that file in parts before acting.

Human version

Make and revise videos from a sourced brief and a reference, with real product assets, a deliberate voice and visible proof. Review the script, the ingredients and the exact final file. Keep earlier cuts and the decisions needed to resume.

---

Agent version

This is the full production sequence used by Studio and customer video hosts. The host supplies its connection instructions; the production decisions and steps live here once.

Read capabilities.md, the host binding and orchestrator.md before starting. The host binding is the entry skill that loaded this package (for GooseWorks: make-custom-video). orchestrator.md is an index; it links one file per group of states. Read each step file when the sequence enters it. A missing capability, approval or dependency blocks that operation; do not invent a tool or a successful status.

Before planning scenes, open the shot-routing guide and the genre card it names for the brief (AI creator UGC, narrator over a montage, motion graphics, hypermotion, chat UI or product demo). Open them again when making or re-rolling a clip. Use Creator Grammar for cuts, pacing, captions and sound.

Before reference study, handoff and final review, read creative contract and review. Before choosing a voice or model, or editing speaking footage, read speech and route qualification. Read product scale and world when product size, use or capacity matters, and libraries, look and rubric for people, sound, camera, light and overlays. Read the editorial review guide before reviewing or revising a cut. Read check severity before acting on any check, review note or redo.

For generated people speaking to camera, read talking-actor production: three new actor options when needed, playable voice auditions before selection, and a qualified speaking route. Other scenes keep their own format and sound. Show ambiguous deliverable choices briefly before paid work; report one short status line per milestone.

The portable scripts give local assembly and measured evidence. scripts/qc_evidence.py fails only a broken final and reports every other measurement as a graded finding; its pass never approves the creative.

For an existing-footage request, use the host's search and real-frame inspection, freeze the original identity, revision, excerpt and audio in the script and ingredients, and follow the same approvals. Read the stored-footage contract in capabilities.md first. A selected original does not authorize buying a generated replacement.