video-score
Skill1 starsChanged 22 days agoMITView on GitHub
Tests a video against a checklist made by viral-checklist and gives a readiness score (0 to 100), the risks that could hold it back, the top fixes, and an estimated chance to beat the channel's usual views (labeled uncalibrated until real results are logged). Use when the user says test this video, score my video, will this go viral, check this video against the checklist, or before publishing.
In this repository10 other files it gives its agents.
- publish-rulerSkill
- reels-researchSkill
- shorts-researchSkill
- similar-videosSkill
- source-checkSkill
- tiktok-researchSkill
- trend-ideasSkill
- video-packagingSkill
- viral-checklistSkill
- youtube-researchSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 0 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
If the user has no file, score only what the link's public page shows and mark the rest cant_tell.
- Very specific: line 62
If Jake Schincariol's chatgpt-youtube-agent-skill is installed, also run its yt-script/hookscore.py --hook "<first 15 seconds of the …
- Very specific: line 66
Run similar-videos to see how the video stands against the top similar ones (not only against the checklist).
- Very specific: line 67
Run video-packaging for titles, description and changes, and publish-ruler for the pre-publish checklist.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video score
Measures a video, looks at it, and checks every checklist item. The result is honest: a readiness score is how well the video follows what works now, not a promise of views.
Inputs
- The video: a local file the user owns. Never download from YouTube, TikTok or Instagram, not even the user's own video, because the platforms' terms do not allow download tools. If the user gives a link to their own video, ask for the file instead:
- YouTube: YouTube Studio, Content, the video's menu, Download (MP4, 720p or 360p, up to 5 times a day per video). Google Takeout exports all of them.
- TikTok and Instagram: the original file from their phone or editor, or the platform's own "download your information" export.
If the user has no file, score only what the link's public page shows and mark the rest
cant_tell.
- A checklist (
checklist.jsonfrom viral-checklist). If there is none, ask for the topic, platform, market and language, and run viral-checklist first. Do not score against a made-up checklist. - Optional: a transcript file, and
results.csvwith the user's past results (see Estimate).
Steps
Measure.
python3 <skill_path>/scripts/inspect_video.py VIDEO --out score-out --every 3 [--transcript FILE]It writes
score-out/metrics.jsonand frames: the hook (0, 0.5, 1, 2, 3 s), one every 3 seconds, and the last frame. Transcription uses--transcriptif given, otherwise faster-whisper if installed, otherwise none (say so in the report).Metrics you can use in checklist
measurefields:metric meaning duration_slength in seconds aspect,vertical9:16,16:9,1:1; true when taller than widecuts_per_minute,longest_shot_s,first_cut_svisual changes, found by comparing frames 5 times a second (crossfades count; changes between two very similar frames can be missed) first_sound_s,first_word_s,words_per_minutewhen sound and speech start, speaking pace (needs word timings) silences_over_1squiet gaps loudness_lufs,true_peak_dbfsloudness and peak (use the platform's current loudness guidance from the checklist, if the research found one) - Look. Open the hook frames, the review frames and the last frame with the image reader. Read the transcript. Note what is on screen: subject in frame 1, text on screen and whether it is readable, captions, faces, brand or credit, end screen space, anything off (garbled text in pictures, watermarks, borders).
- Score every checklist item. For each item write one line:
pass,partial,fail, orcant_tell(when the video or tools cannot show it, like posting time or the title and thumbnail, unless the user gave them)- the evidence: the metric value, the frame time, or the transcript words
- for
partialandfail: the fix, in one plain sentenceavoiditems pass when the video does NOT do the thing. Measured items use the metric, not your opinion.
- Readiness score.
readiness = 100 × sum(weight × points) / sum(weight), with pass = 1, partial = 0.5, fail = 0, andcant_tellitems left out of both sums. Report how many items werecant_tell, because many unknowns make the score less reliable. - Stage scores. Also give the readiness of each stage on its own (same math, only that stage's items): Hook, Story (story, length, pacing), Look and sound (visuals, text, audio), Packaging (title, thumbnail, description), Engagement, Policy. A stage with no scored items shows "can't tell". This shows where the video is weak at a glance (OpusClip shows a similar split: Hook, Flow, Value, Trend).
- Risks. List every failed
mustitem and every failedavoiditem as a risk, most serious first. Policy risks (monetization, copyright, AI labels, made for kids) always go on top. - Estimate.
It returns the chance (in percent) that the video beats the channel's median views. Without 20 logged results it is uncalibrated: it starts at 50% and the readiness moves it at most 20 points. Always print the basis and confidence next to the number. After the video has been out for 7 days, add a row topython3 <skill_path>/scripts/estimate.py --readiness 74 [--results results.csv]results.csv(video,readiness,beat_median) so future estimates learn from real results.
Optional: hook score
If Jake Schincariol's chatgpt-youtube-agent-skill is installed, also run its yt-script/hookscore.py --hook "<first 15 seconds of the transcript>" and add the result as a note. It is a heuristic; its author says a low score is a reason to look again, never a high score a promise.
After the score
- Run
similar-videosto see how the video stands against the top similar ones (not only against the checklist). - Run
video-packagingfor titles, description and changes, andpublish-rulerfor the pre-publish checklist.
Report format
Video score: POC4-atomic-habits-takeaway-1.mp4 (34.5 s, 16:9)
Checklist: book summaries · YouTube long-form · US · en · made 2026-10-23
Readiness: 74 / 100 (20 items scored, 4 can't tell)
Stages: Hook 100 · Story 70 · Look and sound 80 · Packaging can't tell · Engagement 50 · Policy 100
Chance to beat your median views: 59% (uncalibrated, low confidence)
Risks
1. ...
Top 3 fixes
1. ...
All items
pass hook-01 Payoff in the first second ... (frame 0.0 s: Zaid at a huge staircase)
partial pace-02 New visual every 3 to 6 s ... (longest shot 7.0 s at 9.8 to 16.8 s)
...Plain words. Short lines. Never round a score up to make the video look better.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

