ADD THIS SKILL
paste into a new chat
Copy this into a blank Wideframe chat. It tells the agent to build the skill, what to confirm with you first, the steps to follow, and how to verify the result.
PROMPT
I'd like you to add a skill to Wideframe called "label-speakers".
What it should do:
Speaker diarization that maps voices to the real names you supply.
Before you run anything, ask me to confirm:
- Real speaker names + a reference sample per voice (if available)
- Expected number of distinct speakers
- Confidence threshold below which a clip is flagged, not labeled
- Where to write labels: Premiere metadata, sidecar, or both
Then do the work:
1. caltools media-transcript-get-batch with diarization enabled over $1.
2. Cluster voices; map each cluster to a user-supplied name using the reference samples.
3. Any clip below the confidence threshold -> mark "unsure: <best guess>" instead of committing a label.
4. Write labels via caltools prproj edit (metadata) and/or a sidecar speakers.csv.
Do not grade your own work. When the output is ready, hand it to a separate verification subagent that has its own context window. Give it the output, the choices I confirmed, and the pass criteria below. It reviews the quality, writes specific feedback, and returns PASS or FAIL.
Pass criteria:
- Every clip carries a label or an explicit unsure flag.
- Report per-speaker talk time + mean confidence.
- No clip is silently labeled below the threshold.
If it returns FAIL, apply the feedback, redo the affected work, and verify again with a new subagent. Repeat until it returns PASS, then show me the result.
Once it works, save it as a reusable skill named "label-speakers" so I can run it again by saying /label-speakers.RUN IT AGAIN
Once it’s saved, run it on the next project with /label-speakers, or just say:
“Tag every clip with who’s talking. Here are the three names and a voice sample for each.”Run it on your own footage.
7-day free trial. Requires Apple Silicon.
RELATED SKILLS