source-skeleton
Skill1 starsChanged 2 days agoMITView on GitHub
Source skeleton — indexing any incoming material (video, audio, photos, texts) into a reusable timecoded description before any task is applied to it. Use as the first step for any material.
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ai-visualizationSkill
- approval-packSkill
- content-atomizationSkill
- dispatchSkill
- fact-checkSkill
- ffmpeg-skillSkill
- hook-storylineSkill
- idea-diceSkill
- jurySkill
- memory-compactionSkill
- perf-loopSkill
- publish-kitSkill
- seamless-insertSkill
- setupSkill
- shoot-briefSkill
- sound-designSkill
- style-memorySkill
- treg-gatewaySkill
- trend-scoutSkill
- utmSkill
- video-codeSkill
- weekly-cycleSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 31
Write skeleton.json and packed.md and a line in media-index/index.md.
≈ “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
Source skeleton
Rule: first describe what we received, and only then decide what to make of it. A skeleton is made once per file and reused by all videos.
Where it lives
clients/<c>/media-index/assets/<asset_id>/
skeleton.json— the machine description (schema below);packed.md— a compact view for LLM reading (this is the main view; there is no need to read the JSON);frames/— key frames (not committed if heavy; the list of timecodes is kept).asset_id= the first 12 characters of the file's sha1. If it already exists, do not index again.
Order for video
ffprobe: duration, fps, resolution, VFR? → if VFR, convert to a CFR copy in.cache/.- Audio:
ffmpeg -af ebur128(loudness),silencedetect(pauses), clipping. - Word-level transcript (faster-whisper cannot open .mov directly because of the PyAV version:
first
ffmpeg -i in.mov -ac 1 -ar 16000 <abs>/.cache/media/<id>.wav, then pass the model a numpy array; modelsmall, cpu, int8,language=<client language code>,word_timestamps=True): faster-whisper/WhisperX (word_timestamps) or ElevenLabs Scribe (speakers + events). The language is the client'spassport.yaml: language. - Scenes: PySceneDetect (
detect-adaptive). If it is not installed —ffmpeg select='gt(scene,0.3)'. - Key frames: 1 per scene + one every 2–3 s inside long ones. A contact sheet via
tile. - Frame captions: review the frames in batches and describe the shot size (CU/MS/WS/screen/B-roll), what is in frame (object, product, detail, people — who), camera movement, quality (sharpness, shake, light), on-screen text.
- Strong moments: hooks, "wow" frames, complete thoughts, emotions →
highlights. - Write
skeleton.jsonandpacked.mdand a line inmedia-index/index.md.
For photos and images
Caption: what is depicted, the object, the angle, quality, a real photo or a visualization, whether it works as a product reference. For a batch — grouping by object.
For texts and audio
Theses, facts (candidates for product-base — only after the client confirms), quotes with timecodes.
skeleton.json schema
{
"asset_id": "a1b2c3d4e5f6", "source": "yadisk:/phone/IMG_0412.MOV",
"type": "video", "duration": 184.2, "fps": 30, "res": "1080x1920", "vfr": false,
"audio": {"lufs": -21.5, "clipping": false, "noise": "wind"},
"transcript": [{"t0": 12.40, "t1": 12.88, "w": "so", "spk": "S1"}],
"scenes": [{"id": "s07", "t0": 40.1, "t1": 52.6, "shot": "WS", "caption": "empty plot, a fence on the right, the founder points at the spot",
"camera": "static", "quality": {"sharp": 4, "shake": 1, "light": 4}, "people": ["founder"],
"objects": ["plot"], "text_on_screen": null, "usable": true}],
"events": [{"t": 75.0, "type": "pause", "dur": 2.1}],
"highlights": [{"t0": 40.1, "t1": 46.0, "score": 4, "reason": "clean static shot — frame A for f04/f05"}]
}packed.md (example)
# a1b2c3d4e5f6 — IMG_0412.MOV — 3:04, 9:16, 30fps, audio: wind
[000.0–012.3] s01 MS the founder at the gate — "This is the plot where the owners want a garden room…"
[040.1–052.6] s07 WS empty plot, static ★ frame A for an insert
...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.

