check-render
Skill2 starsChanged 27 days agoMITView on GitHub
Gate a rendered video before anyone calls it done - normalize the final mix to -14 LUFS, a code check for silent tails, black and frozen frames, a hook that does not move, subtitles against the spoken words and the on-screen text dose, then an independent frame-by-frame review by the render-auditor agent. Use for "check the render", "is this video ready", "QA this export", "normalize the audio", or before any upload.
- Installs packages
1 other project uses a version of this.
Adopted by
Most recently taken up by
- kanishk-gangwar/filmcraft-mcptheir own version, first seen Nov 8, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- kanishk-gangwar/filmcraft-mcp★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- channel-pulseSkill
- check-thumbnailSkill
- check-voiceoverSkill
- find-topicSkill
- localizeSkill
- make-shortsSkill
- morph-clipsSkill
- produce-videoSkill
- research-factsSkill
- retention-by-sceneSkill
- setupSkill
- upload-privateSkill
- write-scriptSkill
The file
Files
Check the render
Code measures what code can measure; an agent that did not build the video looks at what only eyes can judge.
On drafts and previews run the code gate with --fast only. The full pass, the text dose and the review are for
the final render. Building the render: references/visuals-and-mix.md. What each check measures and how its
threshold was calibrated: references/code-gate.md.
$SKILL below is this skill's folder.
1. Final mix
python $SKILL/scripts/normalize_audio.py render/raw.mp4 --out render/final.mp4Two-pass loudnorm to -14 LUFS with the true peak under -1 dBTP; the video stream is copied and the input is never overwritten. Exit 1 means a target was missed. A "dynamic mode" warning means loudnorm compressed the mix instead of applying one gain: tame the loudest peaks and run it again.
2. Code gate
python $SKILL/scripts/render_check.py render/final.mp4 --srt voice/captions.srt \
--words qa/voice-check-transcript.json [--words-offset 1.5] [--profile live] [--music]One decode pass; writes qa/render-check.json bound to the video's SHA-256.
| Check | Fails when |
| Streams, sound | no video or audio; audio ends 0.5 s early; the last 3+ s are silent |
| Total silence | over 1 s below -60 dB anywhere (a dropout, not a fade) |
| Loudness | outside -20..-9 LUFS (warns outside -16..-12) |
| Black frames | a black stretch of 1 s or more before the final second |
| Frozen picture | 3+ s still in the hook, or anywhere outside the end screen (animated) |
| Hook motion | the first 45 s stand still or only zoom slowly on a still image |
| Subtitles | a cue ends after the video; the file has no readable cues |
| Subtitles vs words | a cue closes over 0.15 s before its last word ends; the median drift is over 0.25 s |
Warnings do not block: body motion outside the band, few cuts in the hook, a number split from its unit, hyphen gaps ("non -cash"), long lines, inputs saved after the render.
--profile livefor camera and screen recordings, where a face or cursor carries the picture: the hook band only warns and body freezes warn at 8 s.--musicfor a music-only soundtrack (no narration): dead air below -50 dB warns at 0.5 s, fails at 1 s.--wordsis the narration's word-timed transcript (check-voiceover saves one);--words-offsetis where the narration starts in the video. Without it, cue ends against speech are not measured.--motion-band EVENTS,FROZEN,SLOWloosens the band for a calmer genre; write down why.
Exit 1 means fix the render.
3. Text dose (long-form, final render)
pip install rapidocr-onnxruntime # optional, only this script needs it
python $SKILL/scripts/text_dose.py render/final.mp4Fails when text cards (large text outside the subtitle band, OCR every 2 s) pass 20% of the runtime, 60% of the
worst 30 s, an 18 s run or 3.5 entries a minute. Writes qa/text-dose.json. Run it on a low-resolution preview
too, before the full render.
When a check is wrong for this video
The creator decides, not the agent, in qa/RENDER-EXEMPTION-<YYYY-MM-DD>.md (Video: line with the hash,
Checks:, Reason:; format in references/code-gate.md). Those FAILs become WARN while the hash matches. Never
add invisible motion or move pixels to slip past a check.
4. Review gate
python $SKILL/scripts/transitions.py render/final.mp4 --frames qa/transition-frames [--cuts 12.5,40,61.2]A review aid, not a gate: marks overlays (label, box, blur) that pop in or out in one frame and hard jumps (only
at --cuts when given) in qa/transitions.json, with before | after PNG pairs.
Give the render-auditor agent the video, the project folder, qa/render-check.json, qa/text-dose.json,
qa/transitions.json and the script. It reuses the measurements and judges frames: empty or broken charts,
clipped text, overlays covering a number or popping without a fade, pictures that contradict the narration, the
end screen.
Save its report unchanged as qa/RENDER-AUDIT-<date>.md. The first lines are fixed:
VERDICT: PASS
Video: render/final.mp4 sha256:<first 12 hex>A re-review is a new file (-2), never an edit. If the video changes after any gate, every gate runs again: the
upload script compares hashes and refuses stale verdicts.
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.