A Claude Code skill that turns a code change, commit range or pull request into a short explainer video, slides and Mermaid diagrams you can send to a client.
You finish a change, run /explain-to-customer, answer four quick questions and get a folder you can send to the client:
- a 30–60 s MP4 with burned-in captions (1920×1080, H.264, small enough for email), plus a poster image and a GIF preview
- 3–5 PNG slides, including a one-page summary card that works on its own in WhatsApp or email
- Mermaid diagrams as SVG and PNG, with the editable
.mmdsource - a short message in the customer's language, ready to paste
The skill reads the diff and the code around it, quotes the product's real UI strings, and can take screenshots of your running app. Those screenshots get numbered callouts and masked personal data, and the "before" shot comes from a temporary git worktree. Every frame, slide and diagram goes through a QA pass before anything is handed over. Your project is only read: nothing in it is edited, committed, stashed or installed.
| shadcn | Whiteboard | Linear |
|---|---|---|
![]() |
![]() |
![]() |
The same change in three styles: switching style re-renders the existing storyboard, no rewrite needed.
21 styles inspired by public design languages, plus Match my project, which builds a style from your repo's own colours, fonts, radius and logo (shadcn/Tailwind CSS variables, tailwind.config, MUI/Chakra/Mantine themes, design tokens, next/font). You can also pass brand hex colours or a website URL.
| Family | Styles |
|---|---|
| Modern SaaS | shadcn · vercel · linear · stripe |
| Clean & friendly | apple · notion · material · fluent |
| Bold & creative | neo-brutalism · glass · whiteboard · terminal |
| Extended | carbon · primer · blueprint · swiss · nord · catppuccin · dracula · clay · executive |
Every style is checked for WCAG AA contrast before rendering, and uses open fonts only, bundled locally with the subsets your language needs.
Requirements: Claude Code, Node.js 22 or newer (older versions fall back to a slower video engine), FFmpeg and git. The GitHub CLI is optional, for PR#123 scopes.
Install it with the skills CLI (the skill is listed on skills.sh):
npx skills add halitsever/explain-to-customer -g -a claude-code-g installs it for your user, so it works in every project; leave it out to install into the current project only. -a claude-code skips the agent picker. Update with npx skills update explain-to-customer -g, remove with npx skills remove explain-to-customer -g.
Manual install with git
# macOS / Linux
git clone https://github.com/halitsever/explain-to-customer ~/explain-to-customer
ln -s ~/explain-to-customer/skills/explain-to-customer ~/.claude/skills/explain-to-customer
# Windows (PowerShell)
git clone https://github.com/halitsever/explain-to-customer "$env:USERPROFILE\explain-to-customer"
Copy-Item -Recurse "$env:USERPROFILE\explain-to-customer\skills\explain-to-customer" "$env:USERPROFILE\.claude\skills\"Update with git pull in the clone (on Windows, copy the folder again).
If ~/.claude/skills did not exist before, run /reload-skills once in Claude Code. New to skills? How to install a Claude Code skill covers both methods step by step.
FFmpeg: brew install ffmpeg, winget install Gyan.FFmpeg or sudo apt install ffmpeg. Without admin rights the skill can install a static build for you.
The first run asks to install the rendering toolchain once, about 250 MB of downloads: npm packages (~151 MB) and a headless Chromium (~99 MB, skipped when Playwright already has it). Everything goes into ~/.cache/explain-to-customer (%LOCALAPPDATA%\explain-to-customer on Windows), never into your project.
/explain-to-customer
/explain-to-customer HEAD~3..HEAD video style=apple out=desktop
/explain-to-customer PR#42 all lang=en --yes
/explain-to-customer "Invoices are now emailed automatically" diagram style=executive
/explain-to-customer image style=#E11D48,#0EA5E9
Without arguments the skill works out the scope itself: everything since your feature branch left the default branch, or your uncommitted changes, or the last commit. It recommends a format, then asks what to create, where to save it, which style to use and whether to take screenshots. Before rendering it shows the storyboard for approval.
| Argument | Meaning |
|---|---|
a1b2c3d, main..HEAD, PR#42, "text" |
What changed |
video · image · diagram · all |
What to create |
style= |
A style id or name, match, brand hex colours or a website URL |
out= |
project (default), cwd, desktop, videos or a path |
lang= |
Language of everything the customer sees; defaults to the language you write in |
audience=technical |
Allows technical terms and code-level diagrams |
--yes |
Skip the approval step |
Output, for example in customer-explainers/2026-11-01-clearer-sign-up-errors/:
clearer-sign-up-errors.mp4 -poster.png -preview.gif
images/01-summary.png 02-….png
diagrams/<name>.svg <name>.png
message.md
_source/ storyboard, scenes, .mmd files, tokens, QA report (for cheap re-renders)
Tweaks after delivery are cheap: ask for a shorter cut, another language or another style, and only the affected outputs are rendered again.
preflight.mjscollects OS, git scope, save locations, a running dev server and tool status as JSON.- Claude reads the diff, writes a change brief and a storyboard, and asks for approval.
theme.mjsturns the style into CSS variables, a Mermaid config and a local font bundle, fixing contrast where needed.capture.mjsscreenshots the app with Playwright.render-diagrams.mjs,render-stills.mjsandrender-video.mjsrender the outputs. Video is a HyperFrames composition with GSAP, rendered locally in headless Chrome and FFmpeg.qa.mjschecks codecs and sizes, extracts frames, flags blank frames, missing glyphs, placeholders, secrets and jargon, and Claude looks at every frame before delivering.
- Secrets, internal URLs, stack traces and real customer data never go on screen; screenshots use demo data and mask emails, phone numbers, tokens and password fields.
- Before delivery, QA scans captions, slide copy, diagram labels and the message for keys, tokens, real emails and internal hosts, then Claude looks at every frame.
- Nothing is invented: no made-up features, numbers or dates, and visuals that are not real screenshots stay clearly illustrative.
- Rendering is local. No cloud rendering, publishing or telemetry.
The skill lives in skills/explain-to-customer/: SKILL.md is the entry point, long guidance is in reference/, styles in styles/, scene templates in templates/scenes/ and the Node scripts in scripts/. docs/ is the website, built with VitePress (npm run docs:dev); its public/ folder also holds the README images and the logo.
cd skills/explain-to-customer
node --test scripts/test/*.test.mjs # save paths per OS, Mermaid label extraction
node scripts/theme.mjs --validate # every style against styles/_schema.json
node scripts/styles-preview.mjs # regenerate styles-preview.png
npx prettier --write "scripts/**/*.mjs" "templates/scenes/*.{js,css}" "styles/*.json"Third-party tools are installed into the tool workspace, not bundled: HyperFrames and Playwright (Apache-2.0), mermaid-cli, culori and rough.js (MIT), GSAP (standard no-charge license), fonts from Fontsource and Selawik (SIL OFL). Remotion is not used because of its company licence. Style names only describe what inspired them; no logos, trademarks or proprietary fonts are included.
Found a bug or have an idea? Open an issue.
Read the contributing guide before sending a pull request.
- claudegrid: opens a grid of Claude Code agents in one terminal window, with a prompt, model and git worktree per pane.
- claude-code-vps: one command sets up an Ubuntu server for Claude Code with Plasma over RDP, hardened SSH, Docker and config sync.
- HaltOS: a monochrome KDE Plasma theme and OS identity layer for Claude Code servers, installed on top of claude-code-vps.
Released under the MIT License.




