craft-prompt
Skill16 starsChanged 4 months agoMITView on GitHub
Turn an approved beat into the exact generation prompt — an exact-seconds constraint prompt for Omni (VFX) or a role-map brief for GPT-Image-2 (graphic). The core value of the kit. Use after GATE 1, per beat.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- sungjunlee/craftkittheir own version, first seen Oct 2, 2026
- V-Songbird/foremantheir own version, first seen Sep 26, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- V-Songbird/foreman★ 35same name
- sungjunlee/craftkit★ 0same 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 repository5 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- analyze-videoSkill
- assembleSkill
- graphic-designSkill
- omni-vfxSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 11
Read prompts/_formula.md §A — it has the full five-block structure + a proven worked example.
- Very specific: line 21
Use the role-map sections from _formula.md (REFERENCE ROLE MAPPING / COMPOSITION / SUBJECT / ENVIRONMENT+STYLE / TEXT RENDERING / QUALITY).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
craft-prompt
The reason this kit beats prompting by hand. Read prompts/_formula.md first — it holds both grammars. One prompt per SEGMENT, not per effect — a segment is a full ≤10s window carrying 2–3 edits, and its prompt gets one TIMING SEQUENCE bullet per edit (see the segmentation rule in the root CLAUDE.md). Then for each approved segment:
If type: vfx → Omni prompt (exact-seconds format)
Read prompts/_formula.md §A — it has the full five-block structure + a proven worked example. The prompt has five labelled blocks — TASK · SCENE CONSTRAINTS · AUDIO · TIMING SEQUENCE · OUTPUT REQUIREMENT — and effects are pinned to exact clip-relative seconds from analysis/words.json. Then:
- Pick the closest recipe in
prompts/recipes/(product-swap, material-morph, set-transform, self-restyle, holograms-vfx, logo-wrap, physics-hook, vfx-sequence-with-character). Adapt it. - TASK — one line naming the edit ("Apply graphic overlays…", "Apply visual transformations…", "Composite animated 3D plush characters…").
- SCENE CONSTRAINTS — "single continuous shot, no scene cuts" + the do-not-change list: face, exact mouth movements, expressions, clothing, room, lighting, camera framing, the foreground RGB microphone, plus any object whose shape must stay. References: attach any inserted character/object/logo as
image_urls(≤7 real PNGs) and say "matching [Image N]"; things already in the shot (bottle→glass) need none. - AUDIO — "Preserve the native source audio track and voiceover identically…" (real audio is re-laid at assembly regardless; submit the unmuted source, never mute).
- TIMING SEQUENCE — one bullet per beat with exact clip-relative seconds read from
words.json(subtract the segment's in-point):- [Xs-Ys]: On the words "<quote>" (spoken A.AA-B.BBs), <event>, anchored to his real motion ("as he raises his hand and points"). Describe events + materials/optics, not adjectives. Many temporally-separated beats in one gen is fine; simultaneous changes are riskier. - OUTPUT REQUIREMENT — where relevant, "final frame clean / 100% source fidelity."
- TEXT TRAP. Omni paints short wordmarks, but never put an action verb or "on-screen text" next to the quoted words — "power on an on-screen text wordmark" leaked "POWER" onto the frame. Say the wordmark "appears / glows brighter" instead. Long/small text still garbles — fall back to a
graphic-designPNG composited with ffmpeg.
If type: graphic → GPT-Image-2 role-map brief
- Use the role-map sections from
_formula.md(REFERENCE ROLE MAPPING / COMPOSITION / SUBJECT / ENVIRONMENT+STYLE / TEXT RENDERING / QUALITY). - Quote every on-screen word verbatim in the TEXT section with placement + colour.
- List the reference images (logos as real PNGs, "use exactly, do not redraw"), ≤16.
- Set aspect = the input video's aspect.
Confirm it's the one that lands
Before writing it back, sanity-check against the recipe's "known failure" notes (e.g. Omni rotation past 180°, text garbling). Write the finished prompt into the beat's prompt cell in beat-plan.md, status → ✍️ prompted. You propose; the user directs. No generation here — that's the next skill, after the user confirms.
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.

