Video Prompt Architect
Use this skill to create final provider-neutral video prompt packs from approved storyboard, motion direction, references, and copy. It prepares prompt artifacts but does not execute video generation.
Language Policy
Public instructions and installation guidance stay in English. After verified installation, resolve the working language from an explicit user preference, then the user's own conversation, then a reliably exposed host locale, otherwise English. Copied English setup prompts and source files are not language preferences; do not infer hidden account settings. Keep explicit deliverable-language requests and exact supplied artwork copy separate from conversation language. Do not translate or rewrite public source files to localize a session.
When To Use
Use this skill when:
- Storyboard, motion direction, references, and copy are ready for video prompt authoring.
- Scene prompts need timing, camera language, motion language, continuity, and expected observables.
- A prompt pack must be reviewed before any user-selected execution path.
Do not use this skill to invent missing structure, choose paid providers, or execute tools.
Inputs
Required:
storyboard_contract: scenes, shot cards, timing, and transitions.motion_direction: pacing, visual system, and attention architecture.reference_pack: continuity anchors and suppression rules.
Optional:
copy_pack: VO, dialogue, supers, captions, and CTA locks.asset_manifest: approved source files or excluded assets.qa_requirements: expected observables and acceptance criteria.target_generator_profile: verified task modes, prompt fields, extension behavior, and negative handling.continuity_carriers: per-shot references, source clips, chained frames, or verified native shared context.
Outputs
Produce a Video Prompt Pack with:
- scene prompts and timing
- camera and motion language
- generation-unit classification and continuity-carrier map
- a revisioned Creative Prompt Contract for every shot with strict locks, timing beats, target adaptation, or execution-readiness needs
- generator format contract and negative handling mode
- copy, VO, or text locks
- constraint and suppression ledger
- expected observables
- QA checks and loopback guidance
Process
- Confirm storyboard and direction are complete.
- Resolve the target generator, task mode, field shape, and negative handling. If unresolved, keep the contract portable and mark formatting pending.
- Classify every request as independent text generation, independent reference-conditioned, frame-chained image-to-video, native extension, or native multishot.
- Map every strict identity, product, location, wardrobe, prop, lighting, camera-side, action-state, and screen-direction lock to a carrier attached to that exact request.
- Convert each shot into a compact contract: duration, frame, subject, one primary action, one primary camera move, setting, lighting, text mode, and end state when it affects the next request.
- Add timing beats only when they fit inside the declared shot duration.
- Convert each shot or scene into standalone prompt-ready language while preserving timing and copy locks.
- Add observable criteria for QA.
- Keep execution, cost planning, and tool choice separate.
Decision Rules
- If storyboard is incomplete, route to
storyboard-architect. - If copy affects timing and is not locked, route to
copy-voice. - Use one standalone prompt per independent scene or shot. A separate request cannot rely on wording from a previous request.
- Keep provider-neutral language unless the user explicitly chooses a specific execution path later.
- Do not output a universal negative-prompt field. Use one only when the verified target interface exposes it.
- If strict continuity is requested without a carrier, route upstream to create the missing reference or label the result approximate with drift risk.
- Treat one primary action and one primary camera move as the default. A more complex unit needs a documented rationale and observable QA evidence.
- Use rewrite-forward only after an actual output is accepted. A storyboard ending or planned frame is not a strict continuation carrier.
- If target-specific syntax is requested, require a current official-source check for the selected surface. Otherwise keep the contract portable.
Guardrails
- Do not run video generation, upload files, or publish outputs.
- Do not invent missing structure, claims, characters, or continuity anchors.
- Do not remove suppression rules.
- Do not hide prompt limitations from QA.
- Do not use
same character,same location,continue previous shot, repeated prose, or a seed as a substitute for a per-request continuity carrier. - Do not imply native text, audio, lip-sync, extension, or multishot behavior before the target surface has been verified.
Handoff
Review gate: promptability_fit.
Hand off to tool-routing-cost or qa-iteration with:
prompt_packtimingcopy_locksgeneration_unitscontinuity_carrierscreative_prompt_contract, revision, and content checksum when usedattachment_planprompt_format_contractacceptance_criteria
QA Checklist
- Every prompt maps to a storyboard scene or shot.
- Timing is preserved and every strict continuity lock has a concrete carrier.
- Every independent prompt is standalone and uses only manifest-valid asset aliases.
- Constraint handling matches the resolved generator profile or is explicitly unresolved.
- Copy or VO locks are explicit.
- Each video contract has observable frame, action, camera, timing, and end-state requirements when continuity needs them.
- Any multi-action or multi-camera exception includes rationale and QA evidence.
- Expected observables are testable.
- No execution path is implied without user choice.