video-edit-harness
AGENTS.md0 starsChanged 41 days agoMITView on GitHub
Work from this checkout with Python 3.11+, uv, FFmpeg, and ffprobe. Run uv sync --locked and use uv run video-harness --help or uv run video-harness session --help for command syntax. Read README.md and the video-editing skill for the editing workflow. Set VIDEO_EDIT_HARNESS_ROOT when invoking the skill from another directory. Keep original media, transcripts, and edit revisions. Derive story spans and omissions from actual transcript word IDs and times. Never invent alignment or treat example IDs as source data. Record the…
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- tiktokSkill
- video-editingSkill
- youtubeSkill
The file
Files
Video Edit Harness contributor instructions
Work from this checkout with Python 3.11+, uv, FFmpeg, and ffprobe. Run uv sync --locked and use uv run video-harness --help or uv run video-harness session --help for command syntax. Read README.md and the video-editing skill for the editing workflow. Set VIDEO_EDIT_HARNESS_ROOT when invoking the skill from another directory.
Keep original media, transcripts, and edit revisions. Derive story spans and omissions from actual transcript word IDs and times. Never invent alignment or treat example IDs as source data. Record the real actor (human, codex, claude_code, or automation) and decision reason. An agent selection is not human approval. A proposed or revised plan requires review. Tie feedback and final review to the exact render SHA-256.
Cloud transcription sends media to OpenAI or Azure OpenAI and may incur charges. Obtain authorization for each source before upload and persist it with session cloud-policy bound to the source SHA and permitted providers. Honor stored deny/unknown on resume. Run doctor before editing; never print credential values. Do not run local Whisper/ASR. Keep credentials out of project JSON, logs, and commits. Do not send messages, publish artifacts, or claim human review on the basis of this file alone; follow the current user's instructions.
Compare color candidates on the same source interval and audio conditions. Preserve natural whites and skin. Static region corrections do not track faces. Generated Apple Log LUTs include conversion to Rec.709; avoid a second FCP Camera LUT. Measure audio before normalization. The -16 LUFS and -1.5 dBTP targets are starting choices. Validate full video/audio decode, source range, frame mapping, and FCPXML rational times. Keep technical checks, synthetic tests, human visual/listening review, FCP GUI import, and distribution playback as distinct evidence. Never auto-adopt a look or remove silence based only on low amplitude. Keep raw logs under output/.
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.

