Add concept-to-storyboard planning helpers

[compoodment]

Problem / opportunity

Hand-authored Vidkit JSON specs are powerful, but they can be verbose when the input is really a loose creative concept: “a fake app tries to approve something, fails loudly, then gets stamped.”

The useful workflow is not asking people to write a storyboard script by hand. The useful workflow is letting a person describe a concept in normal language, then having an assistant or skill turn that into a small, inspectable storyboard plan and finally into ordinary Vidkit scenes/layers/audio/timing.

This issue is about adding a lightweight planning/interchange layer for that assistant-driven workflow. The output should still be readable Vidkit JSON, not a hidden magic blob.

User-facing shape

The primary user-facing workflow should stay concept-first:

Make a short fake application-form video where the form keeps rejecting normal answers, gets stamped NOT NORMAL, and ends with a bonk.

An assistant/skill can then derive a structured plan internally, for example:

{
  "format": "vidkit-storyboard-v1",
  "size": "640x360",
  "beats": [
    {"kind": "template", "template": "application-form", "duration": 1.4},
    {"kind": "beat", "preset": "error_flash", "text": "invalid amount of normal"},
    {"kind": "beat", "preset": "bonk", "text": "NOT NORMAL"}
  ]
}

The structured form is mainly for skills/assistants, reusable examples, and debugging. It should expand to a normal Vidkit spec or render through the existing renderer.

Examples

• A concept prompt that becomes a fake application-form rejection sequence.
• A concept prompt that becomes a fake chat-window interruption sequence.
• A structured storyboard JSON fixture used by an AgentSkill or CLI helper.

Acceptance criteria

• The issue is implemented as an optional assistant/skill-friendly planning layer, not as a required hand-authored user workflow.
• A minimal structured beat format is documented as an intermediate/debuggable representation.
• A command or helper can expand the structured plan to a normal Vidkit spec or render it directly.
• At least five beat kinds or mappings exist, using existing primitives where possible (beat scenes, templates, sprites, shapes, SFX).
• Generated specs are readable enough to debug and modify by hand if desired.
• Storyboard/concept examples render without private assets.

Verification

• validate storyboard/concept examples
• expand at least one structured plan to a normal Vidkit spec
• render at least one example
• inspect contact sheet for beat sequence clarity
• verify audio levels when generated sounds are included
• run python3 tools/vidkit-selftest.py

Non-goals

• Requiring users to write storyboard scripts.
• Full natural-language-to-video generation inside Vidkit core.
• Replacing low-level JSON specs.
• Large cinematic planning system.
• Hiding the generated Vidkit spec from inspection.