Video Generation

# Video Generation

## Use When
- The desired final asset is a generated video or video render plan.
- The request may be a simple clip, storyboard, UGC, podcast clip, reference
  video, talking-head, image-to-video, text-to-video, or research handoff.
- The next decision is task class, request architecture, and execution handoff.

## Do Not Use When
- The user only needs transcription, subtitles, semantic analysis, or edit
  decisioning. Use `media-router`.
- The request architecture is already approved and provider-ready. Use
  `seedance-submitter` or `video-batch-runner`.
- The task is image-only or audio-only.

## Core Boundary
This is the video generation controller. It does not submit jobs.

It must:

1. classify the video task,
2. choose whether references or storyboard must be resolved,
3. hand approved structure to `video-request-architect`,
4. hand submit-ready plans to `seedance-submitter` or `video-batch-runner`.

## Task Classes

| Task class | Use when | Required handoff |
| --- | --- | --- |
| `simple_clip` | one short text/image-to-video clip | `video-request-architect` |
| `storyboard` | panel sequence or contact sheet drives the clip | `storyboard-grid-writer`, then `video-request-architect` |
| `ugc` | creator-style product or app video | `ugc-flow` unless it is one isolated clip |
| `podcast` | voice-led clip, audiogram, or speaker segment | `audio-generation`, then video architecture |
| `reference_video` | benchmark video controls rhythm or motion | `reference-decode` and `reference-contract-builder` |
| `research_handoff` | research/benchmark becomes a generation request | `benchmark-to-brief`, then `video-request-architect` |

## Routing Table

| If not video-generation | Send to |
| --- | --- |
| Understanding or subtitles first | `media-router` |
| Image asset must be created first | `image-generation` |
| Voice or dub must be created first | `audio-generation` |
| Product-to-UGC pipeline | `ugc-flow` |
| Provider-ready request exists | `seedance-submitter` or `video-batch-runner` |

## Three-Layer Boundary
- Controller: classify task and produce the handoff. It must not submit.
- Architect: `video-request-architect` builds provider-agnostic structure. It
  must not call providers.
- Runner: `seedance-submitter` or `video-batch-runner` validates and executes
  normalized requests. It must not make creative strategy decisions.

## Output Shape
Return:

- `taskClass`
- `requiredUpstreamSkills`
- `architectureHandoff`
- `runnerHandoff`
- `referencePolicy`
- `mustNotDo`

## Stop Conditions
- Stop when required user intent, source evidence, or owned input artifacts are
  missing and guessing would change the result.
- Do not ask a runner to infer hook, persona, or reference meaning.
- If an owned CLI or script command fails, report the exact error and stop. Do
  not bypass the failure with metadata-only answers, readiness probing, local
  payload rewrites, fallback providers, or unpublished tools.

## Public Command Boundary

- Choose the smallest matching command or workflow from the user input and run
  it directly.
- If an owned CLI or script command fails, report the exact error and stop. Do
  not bypass the failure with metadata-only answers, readiness probing, local
  payload rewrites, fallback providers, or unpublished tools.
- This public skill is instruction-driven. Produce the controller handoff
  artifact directly from the available evidence.
- Do not call private provider/runtime paths or un