video-qc
Skill0 starsChanged 27 days agoMITView on GitHub
QC a video before anyone sees it: timing, story, stills, audio, frames and device. \"is the cut ready\"
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- cutforgeSkill
- marketing-videoSkill
- media-archiveSkill
- render-and-assembleSkill
- revision-roundSkill
- sound-designSkill
- storyboard-and-lookSkill
- story-designSkill
- video-briefSkill
- video-deliverySkill
- visuals-and-ai-artSkill
- voice-and-narrationSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Run the checks below in order.
- Distinctive: line 27
Run 1-6 on every build; 7 on every cut shown to anyone.
- Distinctive: line 33
Every face that is named: is it that person (facts.md, people map)?
- Distinctive: line 43
Every chapter boundary: is there a bridge?
- Distinctive: line 44
Every name spoken: pronounced as the family says it?
- Very specific: line 57
Write the report to out/qc/QC_vN.md.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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 QC: verify before saying done
"Done" means checked, not rendered. Most of what clients noticed in a real production was detectable from data before anyone watched: wrong people, chronology, a missing relative, abrupt stops, a line running past its shot. Run the checks below in order; each is cheap compared to a client finding it.
The layers
| # | Layer | Tool / method | Blocks delivery when |
| 1 | Timing | python film/timeline.py | any ERROR; unexplained WARN |
| 2 | Story data | python tools/qc/qc_story.py (CHRONOLOGY, CAST, REUSE, RUNTIME) | an asset outside its chapter span, a person the hero names with zero screen time, an unintended reuse, runtime > 5 % off |
| 3 | Facts | facts.md vs every caption, name tag, date note, stamp; sources.md for every quote | any unverified identity, date, ritual name or quote on screen |
| 4 | Picture | python tools/qc/stills.py <chapter> --safe 0.86 and Read every sheet | text or faces outside the safe area, captions < 1/20 height, collisions, wrong person, misspelling, broken conjuncts |
| 5 | Sound | python tools/qc/qc_audio.py; mixer speech-over-bed report | DROP, CUT TAIL, DEAD AIR, loudness > 1 LU off target, narration buried |
| 6 | Render | python tools/qc/verify.py --decode; verify.py --final <master> | frame mismatch, decode errors, duration off |
| 7 | Watch-through | the 540p preview end to end, with sound, at speed | anything that makes you wince; note it with a timestamp |
| 8 | Device | the delivery device in its room (projector, TV, phone with platform UI) | unreadable text, cropped edges, inaudible narration, file will not play |
| 9 | Independent review | film-qc-reviewer agent (and story-editor, compliance-reviewer for brand/regulated work) | any ERROR they confirm |
Run 1-6 on every build; 7 on every cut shown to anyone; 8 before the first screening and the final; 9 on the first full cut, the final, and any regulated deliverable. The reviewer must not be the one who built the cut: it gets the files, not the build history.
What to look for in stills (the sheet labels each frame with film time and shot)
- Every face that is named: is it that person (facts.md, people map)? Background faces are never named.
- Text: spelling in the language, complex-script conjuncts intact, no letter-spacing on Indic text, contrast.
- Safe area box: faces and words inside. Vertical: nothing important under platform UI.
- Collisions: prints over name tags, stamps over faces, captions over prints.
- Continuity: the hero object is the right one in every shot; costumes and places match the date.
- Page turns and transitions: no blank page, no flash frame, no black gap.
How to listen
- Headphones for clicks, cuts and dead air; then the delivery speaker for balance.
- Every chapter boundary: is there a bridge? Every emotional line: is there a breath after it?
- Every name spoken: pronounced as the family says it?
- Music: unbroken inside a chapter, cuts on beats, no swell under words.
The QC report (what the client or the next session reads)
QC v2 · 2026-10-16 · film 19:58 · checks 1-8 run, reviewer pass done
ERROR 8:26 e2/7D name tag says "Grandad" over the grandfather the hero calls "Grandpa" fix: facts.md + shots.json
ERROR 10:42 e2→interval 2.5 s dead air, then a sound cut 0.3 s in fix: bell ring-out + pre-lap
WARN 0:12 open/title line runs 0.44 s past the fixed title shot fix: fixed 5.99 → 6.6
NOTE 12:39 e3/9H print lands under the caption band for 0.5 s fix: printAt 0.6Severity: ERROR blocks delivery; WARN is fixed or recorded as intended in decisions.md; NOTE is optional.
Write the report to out/qc/QC_vN.md; fixes become NOTES_vN rows (see revision-round).
References
references/qc-checklist.md: the full pre-delivery checklist, by layer, with tick boxes- agents:
film-qc-reviewer,story-editor,compliance-reviewer(in this plugin's agents/)
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.