capture-component
Skill0 starsChanged 6 days agoMITView on GitHub
Capture a cropped PNG (or scroll GIF) of any web app UI element for PR screenshots, via browser-harness CDP clip. Use when asked to screenshot/gif a component or element, or produce PR visuals from a running app at any URL.
In this repository7 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- digest-videoSkill
- no-visual-diffSkill
- pr-createSkill
- pr-feedbackSkill
- prior-artSkill
- pr-reviewSkill
The file
Files
capture-component
Crop is done at capture time with CDP Page.captureScreenshot clip (element bbox), not ffmpeg. ffmpeg only stitches GIF frames.
Prereqs
- The target rendered at any URL you can open (local dev server, preview deploy, or a live site).
browser-harnesson$PATH(drives the user's Chrome via CDP).ffmpegonly if making a GIF.
Rules
- Never write captures into the repo. Output to
~/Desktop/<name>-captures/(or ask). - Point at any URL that renders the target in the state you want. If the tooling offers an isolated preview URL (the component alone, no app chrome), prefer it.
- Target the component by an exact selector (
section[aria-label="..."], a unique class). Genericsection/button/ulmay also match page chrome (toolbars, app nav, etc.). cdp()takes params as kwargs, not a dict:cdp('Page.captureScreenshot', format='png', clip=clip).clipfromgetBoundingClientRect(), pad ~2px (rounded corners),scale:2(crisp on retina ≈ 4× output).
Still (PNG)
browser-harness <<'PY'
import base64, os, json, time
out = os.path.expanduser('~/Desktop/CAP-captures'); os.makedirs(out, exist_ok=True)
sel = '.my-component' # exact selector for the component
new_tab('<url>'); wait_for_load(); time.sleep(0.5) # any URL that renders the component
b = json.loads(js("(()=>{const e=document.querySelector('%s');const r=e.getBoundingClientRect();return JSON.stringify({x:r.left,y:r.top,w:r.width,h:r.height});})()" % sel))
clip = {'x':b['x']-2,'y':b['y']-2,'width':b['w']+4,'height':b['h']+4,'scale':2}
shot = cdp('Page.captureScreenshot', format='png', clip=clip, captureBeyondViewport=True)
open(os.path.join(out,'NAME.png'),'wb').write(base64.b64decode(shot['data']))
PYScroll GIF
- Capture N frames stepping the scroll container's
scrollTop0→max, clipping the wrapper each frame.
browser-harness <<'PY'
import base64, os, json, time
out = os.path.expanduser('~/Desktop/CAP-captures/frames'); os.makedirs(out, exist_ok=True)
sel = 'section[aria-label="..."]' # wrapper to clip
new_tab('<url>'); wait_for_load(); time.sleep(0.6) # any URL that renders the component
b = json.loads(js("(()=>{const s=document.querySelector('%s');const u=s.querySelector('ul');const r=s.getBoundingClientRect();return JSON.stringify({x:r.left,y:r.top,w:r.width,h:r.height,max:u.scrollHeight-u.clientHeight});})()" % sel))
clip = {'x':b['x']-2,'y':b['y']-2,'width':b['w']+4,'height':b['h']+4,'scale':2}
N = 12
for i in range(N):
js("document.querySelector('%s ul').scrollTop=%d" % (sel, round(b['max']*i/(N-1)))); time.sleep(0.08)
open(os.path.join(out,'frame_%02d.png'%i),'wb').write(base64.b64decode(cdp('Page.captureScreenshot', format='png', clip=clip, captureBeyondViewport=True)['data']))
PY- Stitch (ping-pong loop + palette for clean colours):
cd ~/Desktop/CAP-captures
ffmpeg -y -framerate 9 -i frames/frame_%02d.png -filter_complex \
"[0:v]split[a][b];[b]reverse[rb];[a][rb]concat=n=2:v=1:a=0,fps=9,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" \
NAME.gifVerify
Readthe PNG / first+last frame to confirm framing and that the scroll actually moved.
PR note
- GitHub PR bodies need images uploaded (drag-drop or
gh); you can't link local files. Leave<img>placeholders or hand the files to the user.
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.

