pr-screenshots
Skill2 starsChanged 2 days agoMITView on GitHub
Attach before/after screenshots to a pull request for any frontend/UI change - cropped to the region of interest, marked with red boxes and arrows, and reviewed by a visual-reviewer agent before the PR goes up. Use whenever raising or updating a PR that touches .vue/.html/.jinja/templates/CSS/Tailwind/theme files, or when asked to "add screenshots to the PR", "show before and after", "crop to ROI", or "review how it looks".
17 other projects use a version of this.
Adopted by
Most recently taken up by
- divinevideo/divine-mobiletheir own version, first seen Nov 16, 2026
- torztomasz/skillstheir own version, first seen Nov 16, 2026
- NickolasBenakis/agent-playbooktheir own version, first seen Nov 15, 2026
- github/squadtheir own version, first seen Nov 8, 2026
- github/awesome-copilottheir own version, first seen Oct 25, 2026
- michalklejdysz/pr-visualstheir own version, first seen Oct 25, 2026
Used with
Projects with it also use security-review.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
SentrySkillsecurity-review
Finds exploitable application security vulnerabilities in code changes. Use for Warden security scans, appsec review, OWASP-style checks, authentication or authorization bugs, injection, XSS, SSRF, path traversal, secrets, unsafe crypto, webhook verification, open redirects, or sensitive data exposure.
Also in 3 projects that use this
Other versions17 projects use a version of this.
Word for word · 1
- bwhtech/skills★ 3100% same
Edited · 0
None found.
Wrote their own version · 16
- github/awesome-copilot★ 41ksame name
- github/gh-aw★ 5.6ksame name
- lukevella/rallly★ 5.4ksame name
- github/squad★ 3.4ksame name
- levy-street/world-of-claudecraft★ 2.3ksame name
- getsentry/skills★ 1.2ksame name
- divinevideo/divine-mobile★ 263same name
- BrOrlandi/my-claude-skills★ 22same name
- nanosite-ai/nanoflow★ 15same name
- NickolasBenakis/agent-playbook★ 5same name
6 more not listed
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 repository9 other files it gives its agents.
- control-roomSkill
- dsa-conscious-codingSkill
- excalidraw-architectureSkill
- improve-codebase-architectureSkill
- issue-to-phasesSkill
- raise-prSkill
- visual-reviewerSkill
- wayfinderSkill
- workSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.
0
Common
Most of the 15 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Use the run skill (or the project's own run instructions) to get the app up, and agent-browser to drive it.
- Distinctive: line 54
Use the same crop box for the before and after pair so they line up.
- Very specific: line 69
Run the visual-reviewer skill (visual/UX judgement, reads the images), plus a code reviewer for the frontend diff, before the PR body is …
- Very specific: line 84
Use the house convention: commit the screenshots into the PR branch under docs/implementation/media/ and link them as blob URLs with ?raw=1:
- Very specific: line 93
Only when images must outlive the branch: with a GitHub session in agent-browser, upload into the PR comment box's hidden input[type=file] …
“Common” is measured: kept by at least 6 in 10 of the 15 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
PR screenshots
A frontend PR without pictures cannot be reviewed. Reference style: frappe/frappe#35626.
When this applies
The diff touches any of: *.vue, *.html, *.jinja, templates/, themes/,
*.css, Tailwind classes, or frontend *.js/*.ts that changes what renders.
Pure backend / test / config diffs skip this entirely.
1. Run the app and capture the pair
Use the run skill (or the project's own run instructions) to get the app up, and
agent-browser to drive it. Capture the same route, same viewport, same data for
both states — only the code differs.
# after (current branch)
agent-browser --session pr open "<url>" && agent-browser --session pr wait 3000 \
&& agent-browser --session pr screenshot <scratchpad>/<view>-after-full.png
# before — temporarily restore the base version of just the changed files
git checkout <base-branch> -- <changed files>
# wait for HMR/rebuild, then re-open the same url
agent-browser --session pr screenshot <scratchpad>/<view>-before-full.png
git checkout <this-branch> -- <changed files> # ALWAYS restoreRules:
- One pair per distinct view the change touches (list page + detail page + mobile width if the change affects narrow layouts).
- Seed or navigate to real data — an empty state proves nothing.
- Restore the working tree before doing anything else; verify with
git status. This matters most on a shared server checkout, where another session may commit leftovers. - A pure-addition change has no "before" — ship a single "after" image instead of faking one.
- The
-full.pngcaptures stay in the scratchpad; only the annotated crops (step 2) ship. - Toasts and error pop-ups do not render in a headless capture. Assert the message from the DOM instead, skip that image, and quote the error text in the body. Never fake it.
- Alpine
x-showlooks frozen headless (rAF never drains): shimwindow.requestAnimationFrame = cb => setTimeout(() => cb(performance.now()), 0)before driving the page.
2. Crop and annotate
Crop each capture to the region that changed, scale it 2x, and mark the change with a red rounded box, plus an arrow when the eye needs leading (e.g. "click this → that opens"). An arrow must run through empty space; if it would cross another field or text, use the box alone. Use the same crop box for the before and after pair so they line up.
# needs Pillow; without it: uv run --with pillow python ...
python3 <this skill's folder>/annotate.py <scratchpad>/header-after-full.png \
docs/implementation/media/header-after.png \
--crop 900,0,1280,130 --box 1142,18,1178,62 --arrow 1070,118,1142,58Coordinates are source-screenshot pixels: measure them with
getBoundingClientRect() in agent-browser eval rather than guessing. Read the
annotated image before using it — a box that misses its target is worse than none.
3. Get the screenshots reviewed
Run the visual-reviewer skill (visual/UX judgement, reads the images), plus a code reviewer for the
frontend diff, before the PR body is written. Give
the reviewer the image paths, the intent of the change, and the running URL, and tell
it explicitly not to launch or modify anything.
Act on REQUEST CHANGES first, then re-capture the "after" images. Do not publish a PR carrying screenshots the reviewer rejected, or one that shows an unrelated bug (a wrong total, a stray error) the PR does not fix — crop it out or drop the image.
4. Publish the images
GitHub's drag-and-drop upload host (github.com/user-attachments/assets/<uuid>) is
browser-session-only — there is no API for it and gh cannot use it. Relative repo
paths do not render in PR/issue bodies either.
Use the house convention: commit the screenshots into the PR branch under
docs/implementation/media/ and link them as blob URLs with ?raw=1:
https://github.com/<owner>/<repo>/blob/<branch>/docs/implementation/media/<name>.png?raw=1Name files after the change, not the PR number: <view>-before.png / <view>-after.png. The URL dies with the branch after merge; that is accepted.
Only when images must outlive the branch: with a GitHub session in agent-browser, upload
into the PR comment box's hidden input[type=file] and use the user-attachments URL it inserts.
5. Put them in the PR body
The intent must be clear at a glance: the problem in user terms, what you can do now, where to see it, then the images. Nothing else — no "Verified:" line, no implementation notes (file names, routes, mechanisms), no test logs, no "Generated with Claude Code" / session-link footer. Those are noise; reasoning belongs in code comments or the review thread.
One image per row, never a side-by-side table. A capture squeezed into half a table column is unreadable.
1. <the problem, in user terms>
2. <what you can do now>
You can visit <route / screen> to see it.
## Before
<img src="<url>/<view>-before.png?raw=1" width="760">
## After
<img src="<url>/<view>-after.png?raw=1" width="760">
Set width to the annotated image's pixel width, capped at 900. The 2x scale then keeps
it sharp without rendering as a giant column.
Then verify: curl -sL -o /dev/null -w "%{http_code} %{content_type}" <raw-url> must
print 200 image/png for every image (?raw=1 redirects, hence -L), and the body must
come back changed when read again —
gh pr edit fails silently on repos with Projects (GraphQL scope), so patch with
gh api -X PATCH repos/<owner>/<repo>/pulls/<n> -f body="$(cat body.md)" and read the
body back.
Checklist before you say it's done
- before + after captured at the same route/viewport/data
- working tree restored (
git statusclean of the temporary checkout) - each image cropped to its ROI, change boxed in red, annotated image looked at
- visual reviewer ran; its blocking findings are fixed and re-captured
- every image URL returns 200 with
content_type: image/png - PR body read back: problem / what you can do now / where to see it, images render, no Verified line or footer
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.

