跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

shotkit

Run the whole Shotkit pipeline from one line of brief and open the board. Use when the user types /shotkit, or asks to "plan a video", "storyboard this and show me", "make a shot plan with prompts", or wants a storyboard, the prompts for their image or video generator and one page to see it all on. Also use when the user points at a folder of their own photos and clips, or says "use my footage", "make a video from these", "what can we make from this shoot". Plans the shots, writes the prompts, renders the board and opens it in a browser. For one step only, such as prompts for an existing shots.json or a check of one frame, use that step's own skill.

whystrohm · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add whystrohm/shotkit --skill shotkit

技能说明


name: shotkit description: Run the whole Shotkit pipeline from one line of brief and open the board. Use when the user types /shotkit, or asks to "plan a video", "storyboard this and show me", "make a shot plan with prompts", or wants a storyboard, the prompts for their image or video generator and one page to see it all on. Also use when the user points at a folder of their own photos and clips, or says "use my footage", "make a video from these", "what can we make from this shoot". Plans the shots, writes the prompts, renders the board and opens it in a browser. For one step only, such as prompts for an existing shots.json or a check of one frame, use that step's own skill.

Shotkit

One line of brief in. A plan, prompts and an open board out.

This skill runs the other four in order and puts the result on screen. The user should not have to know the skills' names, ask for each step, or go looking for files. They type a brief and a browser tab opens with every shot on it.

What the user sees

  1. They type a brief, for example /shotkit 30-second founder explainer for Harbor Roast, 9:16.
  2. About two minutes later a browser tab opens: the board, with every shot on a timeline and a contact sheet.
  3. The board fills in with a prompt for each shot.
  4. Your reply is short: what was made, what you assumed, and the one next step.

Where the tools are

Use the first of these that exists, and write it as SK below:

Installed howSKBrand packs
./install.sh~/.claude/shotkit-tools/sk~/.claude/shotkit-brand-packs/
./install.sh --project./.claude/shotkit-tools/sk./.claude/shotkit-brand-packs/
As a plugin, or from a clonetools/sk, two folders above this skill's own folderbrand-packs/, beside tools/

Where another Shotkit skill writes ~/.claude/shotkit-tools/sk or ~/.claude/shotkit-brand-packs/, use the row that matches this install.

If none exists, the skills were copied without the tools. Tell the user, and offer to run git clone https://github.com/whystrohm/shotkit.git && cd shotkit && ./install.sh.

SK needs no setup: the checks make their own Python environment the first time they run.

Steps

1. Read the brief and fill the gaps yourself

Take from the brief: what the video is for, who it is for, the length, the aspect ratio, the brand, and the generators the user names.

Do not stop to ask for what a sensible default covers. State each default in your final reply so the user can change it.

MissingUse
Length30 seconds
Aspect9:16 when the brief says Reels, Shorts, TikTok, Stories or "vertical"; otherwise 16:9
Frameworkthe one storyboard-architect would pick for this brief
Voiceoveron-screen text only, unless the brief mentions a speaker or a script
GeneratorsFlux for stills and Kling for motion

Ask one question, and only one, when the brief does not say what the video is about.

2. Settle the brand

In this order, first match wins:

  1. A brand lock the user names, or a brand-lock.md or brand/*.md in the project.
  2. A site URL, a PDF or screenshots in the brief: run brand-lock-extractor on them and save the result as brand/<name>.brand-lock.md.
  3. A pack named in the brief, from ~/.claude/shotkit-brand-packs/.
  4. None of these: use ~/.claude/shotkit-brand-packs/_starter.md. Set brand_lock_configured: false in run.json. Do not invent a palette for a real company, and do not put the company's name on a palette you made up.

3. Plan the shots

Follow storyboard-architect and write its five files into ./output/. If ./output/ already holds a run, write to ./output-<short-name>/ instead and say so. Never overwrite a run.

Then check them:

SK validate output/

Fix what it reports and run it again until it passes.

4. Open the board

Start the live board in the background, so it stays up while you work and reloads by itself as files land:

SK board output/ --serve --open

Run that as a background command. If you cannot run background commands, write the page and open it instead:

SK board output/ --open

Tell the user in one line that the board is open.

5. Write the prompts

Follow visual-prompt-forge for the generators from step 1. Then:

SK validate output/
SK board output/

A live board reloads on its own. The second command keeps output/board.html current for when the live board is closed.

6. Report

Keep it short. The board carries the detail.

  • One line: what was made and where (output/, and the board address or path).
  • The defaults you used, as a short list.
  • The next step, which is the one the board shows: make the frames from the prompts. Say the three ways: paste each prompt into a generator, drop frames on the live board, or say "make the frames on fal" to have Shotkit make them with their fal key.

Starting from the user's own photos and clips

Use this route when the brief names a folder, when files were dropped on the board (output/media/media.json exists), or when the user says the footage is theirs. The plan is built around what they have. A generator only fills what is missing.

  1. List it.
    SK media scan <their folder> --out output/
    
    Their files are never moved or changed. Each one gets an id (m01, m02) and a thumbnail in output/media/thumbs/. A clip gets three: first, middle and last frame.
  2. Look at every file. Read each thumbnail with the Read tool. Do not plan from file names. For each item write one or two plain sentences into its note in output/media/media.json: what is in it, where the subject sits, what in it is usable and what is not (soft focus, a logo, a face, shake, bad light). For a clip say which seconds are the good ones. Describe only what you can see.
  3. Plan around it. Follow storyboard-architect, with these additions in `s
阅读完整来源 ↗