Discover Moments

Purpose

Find candidate short-form moments from a timestamped transcript during an Opal Job.

Install from SKLx

This skill is published on SKLx and can be installed locally or on an Opal server:

skl add "discover moments" --registry https://sklx.one

Current API

Moment discovery is an internal job phase, not a standalone public endpoint.

Use:

POST /api/upload
GET  /api/jobs/:jobId
POST /api/jobs/:jobId/cancel

Inputs

{
  "jobId": "string",
  "transcriptSegments": [
    {
      "start": 0,
      "end": 0,
      "text": "string"
    }
  ]
}

Outputs

Candidate moments are persisted on the job after ranking/review preparation.

{
  "jobId": "string",
  "status": "ready_for_review",
  "clips": []
}

Steps

1. Split transcript into chunks.
2. Send chunks to the configured Ollama model.
3. Find strong standalone moments.
4. Pass candidates into ranking.
5. Persist final Sparks on the job.

Rules

Return JSON only from model calls. Keep clips between 20 and 90 seconds. Prefer natural starts and endings. Avoid duplicate moments. Do not invent timestamps.