video-reviewer

name	video-reviewer
description	Review a previous short-form vertical video and plan the next episode. Use when the user asks to review yesterday's video, critique a Reel/TikTok/Short, build a video review pack, extract stills/contact sheets, analyze captions/pacing/cover/CTA, or turn prior post performance into a next-day shoot plan.

Video Reviewer

Review the previous episode as evidence, not vibes. Build a review pack first, inspect the generated artifacts, then write a critique and a next-episode brief.

Core workflow

1. Identify the day to review.

   • If the user says "yesterday" or "previous day", infer the latest completed edits/day_NN/final.mp4.
   • Use repo paths first: edits/day_NN/final.mp4, edits/day_NN/final_captioned.captions.json, published/day_NN/receipt*.json, and series/days/NN-*.md.

2. Run the review pack script:

   python3 plugins/day-02-video-reviewer/skills/video-reviewer/scripts/build_review_pack.py NN
   

3. Inspect the artifacts in .context/reviews/day_NN/.

   • Open hook_sheet.jpg, contact_sheet.jpg, caption_sheet.png, and any cover frame.
   • Read review_pack.json for specs, caption page counts, receipt URLs, and audio levels.
   • Read the script/transcript/caption files when available.

4. Write series/days/NN-review.md.

5. Write series/days/MM-brief.md for the next day.

Do not publish or delete posts during review. If a live post has a serious issue, recommend the action and wait for explicit approval.

Review rubric

Use timestamped evidence wherever possible.

• Hook: Does the first 3 seconds establish format, tension, and stakes?
• Proof: Are claims supported visually instead of only asserted?
• Pace: Are there pauses, confusing transitions, or overly fast reads?
• Captions: Are captions readable on mobile, short enough, timed to speech, and clear of faces/platform UI?
• Visual variety: Is there enough movement between selfie, screen, b-roll, and proof?
• Audio: Is loudness clear, clipped, too quiet, or noisy?
• Cover: Does the cover read at feed size and make a viewer want to tap?
• CTA: Is the comment/DM action specific and worth doing?
• Series continuity: Does the episode set up tomorrow?
• Pipeline learning: What should become more automated next?

Output format

For NN-review.md:

# Day NN Review

## Verdict
- Overall: N/10
- Keep:
- Fix next:

## Evidence
| Area | Timestamp | Observation | Decision |
|---|---:|---|---|

## Scorecard
| Dimension | Score | Notes |
|---|---:|---|

## Pipeline Improvements

## Metrics To Add Later

For MM-brief.md:

# Day MM Brief

## Angle

## Hook

## Story

## Shot List

## Pipeline Build

## CTA

## Success Criteria

Judgment rules

• Prefer concrete fixes over broad advice.
• Preserve what worked before changing style.
• Make tomorrow's episode a visible build step, not just a status update.
• If metrics are missing, produce a creative review anyway and list the metrics to backfill.
• If the video is already live, treat deletion/repost as a last resort.