ccvd
/ccvd
Use when comparing how a web page or site looks before and after a change (two commits, branches, deployments or URLs), when making before and after screenshots for a pull request, or when checking a UI change or refactor for visual regressions.
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 5
No signals detected. Repository changed on 29.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Use when comparing how a web page or site looks before and after a change (two commits, branches, deployments or URLs), when making before and after screenshots for a pull request, or when checking a UI change or refactor for visual regressions.
Original author description
Use when comparing how a web page or site looks before and after a change (two commits, branches, deployments or URLs), when making before and after screenshots for a pull request, or when checking a UI change or refactor for visual regressions.
How it works
- 1. Get two addresses
- 2. Live view for the user
- 3. Images for a PR, or to check your own work
- 4. Clean up
- Common mistakes
SKILL.md preview
--- name: ccvd description: Use when comparing how a web page or site looks before and after a change (two commits, branches, deployments or URLs), when making before and after screenshots for a pull request, or when checking a UI change or refactor for visual regressions. --- # ccvd: Claude Code Visual Diff ccvd renders two versions of a site in one hidden Chromium and compares them pixel by pixel. Changed pixels are red; everything else is shown faded. It has two modes: - **Live view**, for the user: before, diff and after side by side in their browser, with scrolling, clicks, typing and links mirrored to both sides. - **shot**, for you: before, after, diff and composite PNGs per page and size, with the count of changed pixels. Run it as `npx -y @laveez/ccvd@0.1.2`, with the before address first and the after address second. It needs Google Chrome, or the Chromium from `npx -y playwright-core install chromium`. ## 1. Get two addresses ccvd compares two running sites. Use deployed URLs as they are (production against a local dev server works too). To compare two git versions of the project: 1. Check out each version into its own folder outside the working tree: `git worktree add --detach <dir> <ref>`, or `git archive <ref> | tar -x -C <dir>` when the build does not need git. 2. Install dependencies in each folder if the project needs them. 3. Start the project's own dev server in each folder, in the background, on two free ports on 127.0.0.1, and note how to stop each one (its process id, or your tool's background task). Find the command in the README or `package.json` scripts. Examples: `npm run dev -- --port 8101`, `php -S 127.0.0.1:8101`, `python3 -m http.server 8101 --bind 127.0.0.1 --directory <dir>`. 4. Wait until both servers answer before running ccvd. ## 2. Live view for the user Run in the background; it keeps running until stopped: ```bash npx -y @laveez/ccvd@0.1.2 http://127.0.0.1:8101/ http://127.0.0.1:8102/ ``` It opens the view in the user's browser and prints its address (`Open http://127.0.0.1:2283/?t=…`). Give the user that address. Add `--no-open` when the user is on another machine. Useful options: `--size desktop|laptop|tablet|phone|WIDTHxHEIGHT`, and `--no-reduced-motion` when the user wants to see the pages' animations (reduced motion is on by default, so animated pages hold still). ## 3. Images for a PR, or to check your own work ```bash npx -y @laveez/ccvd@0.1.2 shot http://127.0.0.1:8101/ http://127.0.0.1:8102/ --path / --path pricing.html --size desktop --size phone --out <dir> --json ``` - Take each `--path` from the site's own links: `pricing.html` on a plain static server, `/pricing` on most frameworks. It resolves against each address like a link. Without `--path`, both addresses are captured as given, query and hash included. - Sizes: `desktop` 1920x1080, `laptop` 1440x900, `tablet` 768x1024, `phone` 440x956, or `WIDTHxHEIGHT`. Full page by default; `--viewport` for the visible part only. - Files are `<dir>/<page>-<WIDTH>x<HEIGHT>-<kind>.png`. The page is named after its path (`/` is `index`), and kind is `before`, `after`, `diff`, `pair` (before and after, labelled) or `triple` (before, diff and after). For a PR, attach the `-pair.png`, or `-before` and `-after` side by side in a table. - `--json` prints to stdout, per page and size: `files` (absolute image paths), `changed` and `total` pixels, and each side's `url`, `status` and `error`. Then, before reporting: - Open each `-diff.png` next to its `-pair.png`: the diff shows where pixels changed, the pair shows what changed. Describe every change and say whether it was intended. - `changed: 0` means the pages are pixel-identical, which proves a refactor changed nothing visible. - A side with `status` 400 or higher, or an `error`, did not load. Fix the path or the server and run again; a diff against an error page means nothing. - A layout shift turns everything below it red. Find where the red starts, then compare the area below it in the pair image to see whether anything besides its position changed. ## 4. Clean up `shot` exits by itself. Stop the live view and the servers you started, then check the ports are free: stopping `npm run dev` can leave the server it started running. Remove the folders (`git worktree remove <dir>`). ## Common mistakes | Mistake | Fix | |---|---| | Copying `--path` values from an example | Use the paths from the site's own links | | Diffing while a dev server still compiles | Wait for its ready line, then run ccvd | | Both servers on one port | One free port each | | `--no-reduced-motion` without a reason, so animations keep the diff from settling | Leave reduced motion on unless the animation itself is under review | | Reporting the image count, not what changed | Describe each change, using the diff and the pair |
Versions
The database has no other skills named ccvd.
Database snapshot: 13.11.2026 · package version 7b9428c40f6d (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90capture-screenshots
Capture screenshots of each state, size and setting as evidence for a PR. Use for each UI change. - 100visual-regression
Catch unintended visual changes by comparing screenshots against a baseline. Use after any change to a web UI, before calling it done, and whenever a change claims to be visual-only. Web UI only; needs a browser. Only invoke via the orcinus router. - 100sideshot
Before-and-after screenshots of the part of a page a pull request changes (a component, a section, a style), not the whole page, on a desk and a phone, optionally in dark mode, with what changed marked, shown to the user first, then kept on a branch of their own and put in the… - 95before-and-after
Add existing screenshots or screen recordings to a GitHub pull request as a before/after or preview block. Use when a PR needs visual media attached to its description. Browser navigation and capture belong to agent-browser. - 100dev-visual-verdict
Visual regression testing — capture before/after screenshots of UI changes and compare. Use when a UI change might have unintended visual impact across components. - 100recording
Record evidence of a web app — an operation video, screenshots and a runbook — for a pull or merge request, a report, or a hand-off. Use when someone asks for proof that a change works, hands over an MR/PR link, or describes a page and the steps to capture on it.