video-factory
AGENTS.md67 starsChanged 17 days agoMITView on GitHub
An AI video studio run by a coding agent. You write and direct; engine/ talks to Kie.ai (images, video, music, vision judge) and ElevenLabs (voices, sound effects); the user approves two gates. First time here? If studio.json doesn't exist at the repo root, follow SETUP.md, then the onboarding skill (.claude/skills/onboarding/SKILL.md). Every request: read studio.json (the creator's profile), then .claude/skills/studio/SKILL.md (the pipeline and its non-negotiable safeguards), then the vertical DNA file it points to. Skills are plain Markdown in .claude/skills/*/SKILL.md: if…
- Reads credentials
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- castSkill
- cinematographerSkill
- format-longformSkill
- format-seriesSkill
- onboardingSkill
- referencesSkill
- story-architectSkill
- studioSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 5
First time here?: If studio.json doesn't exist at the repo root, follow SETUP.md, then the onboarding skill …
- Very specific: line 7
Every request: read studio.json (the creator's profile), then .claude/skills/studio/SKILL.md (the pipeline and its non-negotiable …
- Very specific: line 16
Watch the whole final: before delivering: studio.mjs qc (identity across every cut, transcript, format, loudness, contact sheet), then …
- Distinctive: line 18
Likeness and IP: only the user's own face and voice (with consent) for "clone me".
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md 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
AI Video Factory: instructions for agents
An AI video studio run by a coding agent. You write and direct; engine/ talks to Kie.ai (images, video, music, vision judge) and ElevenLabs (voices, sound effects); the user approves two gates.
First time here? If studio.json doesn't exist at the repo root, follow SETUP.md, then the onboarding skill (.claude/skills/onboarding/SKILL.md).
Every request: read studio.json (the creator's profile), then .claude/skills/studio/SKILL.md (the pipeline and its non-negotiable safeguards), then the vertical DNA file it points to. Skills are plain Markdown in .claude/skills/*/SKILL.md: if your agent doesn't load them automatically, open and follow them yourself.
Rules
- Flow v5 is the only path. Every video starts with
node engine/studio.mjs new(the 7-question intake) and moves only bynode engine/studio.mjs status <slug>. Approvals record the user's own words (studio.mjs approve <slug> <step> --said "..."). The engine refuses any step out of order; a Claude Code hook (.claude/settings.json) denies one-off scripts that call Kie or ElevenLabs and any video sent without a delivery receipt. - Never spend past the approved budget. Approving the Production Book locks it in code;
gen.mjsrefuses any job past it, and any video job before it. Any extra spend needs the user's explicit yes, thennode engine/phase.mjs budget <slug> --add N --said "...". - Creative calls are the user's. Captions, zoom, crops, cuts, music, voices, look: offer options, wait for their choice.
- The user's gates: concept, script, cast, the Production Book (locks the budget) and the retro; plus the series bible for a new series. Nothing paid in video happens before the book.
- Best quality for the lowest cost, from live prices. No video model is excluded by rule. Before every Production Book,
node engine/studio.mjs prices <slug>analyzes Kie's live prices and recommends the best quality that fits the budget; the flow won't build the book without it. Images: GPT Image 2.5 (gpt2.5). - At most one reroll per scene, and only inside the budget. Never change the engine in the middle of a production: deliver, report, fix after.
- Watch the whole final before delivering:
studio.mjs qc(identity across every cut, transcript, format, loudness, contact sheet), thenstudio.mjs deliver. Never hide a known problem. - Keys live only in
.env. Never print them, never put them in commands, commits or URLs. - Likeness and IP: only the user's own face and voice (with consent) for "clone me". Never real public figures, copyrighted characters or brands the user doesn't own. Never name real shows or actors in prompts.
- Reply in the user's language.
Formats and where they live
| The user wants | Skill / DNA |
| Any video (entry point) | studio |
| UGC ad | studio + verticals/ugc-ad.md |
| Micro-drama / vertical novela series | format-series + verticals/micro-drama.md |
| Sitcom or comedy series | format-series + verticals/sitcom-90s.md + sitcom-writing-bible.md |
| Fantasy short | studio + verticals/fantasy-short.md |
| Meditation, sleep, ambient, focus (10 min to 3 h) | format-longform + verticals/meditation-longform.md |
| AI influencer / clone me | studio + cast + verticals/influencer-talking-head.md |
| Anime, cinematic, podcast clip | studio + the matching DNA |
| Scripts and jokes | story-architect |
| Shot lists and continuity | cinematographer |
| Characters, sets, voices | cast |
Engine map
- Flow:
studio.mjs(intake, status, approvals,accept-take,override, finalqc,deliver),lib/flow.mjs(recipes,requireStep),phase.mjs budget,radio.mjs(radio cut + pacing QC),book.mjs(Production Book, every format). - Generation:
gen.mjs(all paid generation; budget guard injob()),series.mjs(series intro),cast.mjs,style.mjs. - Evaluation:
continuity.mjs(vision judge;--identity,--jumps),lib/identity.mjs(faces and marks against the locked face anchor),lib/jumps.mjs,lib/dialogue.mjs,qc.mjs. - Editing:
edit.mjs(scenes formats),cut.mjs(clips formats: influencer, UGC, cinematic...),longform.mjs(long-form ambient),animatic.mjs. - Data:
bible/(characters, sets, styles, sound library),series/<id>/,projects/<slug>/(shotlist, ledger, renders),templates/(starting points). - Money: 1 Kie credit = $0.005. Each project's spend is in
projects/<slug>/ledger.json. - Checks:
npm run doctor(machine),npm run check(offline smoke tests and a secret scan).
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.

