version: "1.0.0" name: segment-storyboard description: Extract at least one storyboard frame per video shot segment so later OCR and segment evidence fusion can cover gaps missed by global sampling.

Segment Storyboard

Use this after video-shots when global storyboard sampling misses many shot segments or segment.evidence.json reports too many empty segments.

This skill owns:

• reading video.shots.json
• extracting one to three frames per shot
• writing a standard storyboard.manifest.json under segment-storyboard/
• making the output reusable by storyboard-ocr, storyboard-transitions, and

segment-evidence

Inputs

Use the same bundle locators as video-artifact-intake:

{
  "outputDir": "./output/example-run",
  "framesPerSegment": 1,
  "format": "jpg"
}

framesPerSegment defaults to 1 and is capped at 3. storyboardOutputDir is optional and defaults to ./segment-storyboard inside the run directory.

Outputs

The result contains manifestPath and manifest. The manifest is compatible with storyboard-ocr; each frame includes sourceShotIndex and segmentPosition for traceability.

Sequencing Guidance

Run:

1. video-shots
2. segment-storyboard
3. storyboard-ocr with storyboardDir set to the generated

segment-storyboard/ directory

1. optionally storyboard-transitions
2. segment-evidence

When both storyboard/ and segment-storyboard/ exist, intake prefers segment-storyboard/ for review because it gives every shot a coverage chance.

Interpretation Notes

This improves coverage, not understanding. It gives OCR and reviewers a frame inside each shot, but static frames can still miss motion, clicks, narration, or short-lived UI states.

Repo-side runner:

node --import tsx scripts/harness/segment-storyboard.ts

Installed-pack runner:

node ./.video-evaluator/agent/run-tool.mjs segment-storyboard