Install
$ agentstack add skill-alexanderop-app-screenshots-app-screenshots ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Annotated Screenshot Documentation
Generate a markdown document with annotated screenshots of any web app or live website.
Prerequisites
agent-browserCLI installed globally (npm i -g agent-browser && agent-browser install)
Phase 1: Determine the Target
A) Live website
- Normalize URL (add
https://if missing):
``bash agent-browser open https://example.com && agent-browser wait --load networkidle ``
- Dismiss cookie/consent banners using semantic locators (most reliable):
``bash agent-browser find text "Accept" click ` Or snapshot to find the accept button ref, then click the ref directly: `bash agent-browser snapshot -i -s "[role=dialog], [class*=cookie], [class*=consent]" -c agent-browser click @e1 # Use the ref from snapshot ` Verify it's gone: agent-browser is visible "[role=dialog]" or agent-browser wait "[role=dialog]" --state hidden`
- For login walls or anti-bot issues, see [references/live-sites.md](references/live-sites.md).
B) Local dev server
- Find or start the dev server (check ports 3000/5173/4321/8080, or start via
package.json) - For self-signed certs:
agent-browser --ignore-https-errors open https://localhost:
Phase 2: Map the Site
- Open homepage:
agent-browser open && agent-browser wait --load networkidle - Take a quick auto-annotated screenshot to survey the page — this labels all interactive elements with numbered refs:
``bash agent-browser screenshot --annotate docs/screenshots/_survey.png ` Read the output legend to understand what's on the page (buttons, links, inputs mapped to @e` refs).
- Snapshot navigation to discover pages:
``bash agent-browser snapshot -i -s "header, nav, [role=navigation]" -c ` For deep DOM: agent-browser snapshot -i -c -d 3`
- For SPAs with client-side routing: also check route definitions in the source code (
grep -r "path:" src/routeror similar) since not all routes may be linked from the nav. - Ask the user to confirm the page list. For live sites, suggest 5-10 pages max.
Phase 3: Screenshot Each Page
Workflow per page: validate selectors → annotate → check for FAILED → verify screenshot → fix or continue.
Setup (run once before all screenshots):
agent-browser set viewport 1440 900
export AGENT_BROWSER_SCREENSHOT_DIR=docs/screenshots
Block ads and tracking scripts that create overlays or slow page loads (live sites):
agent-browser network route "**/*.doubleclick.net/**" --abort
agent-browser network route "**/googlesyndication.com/**" --abort
agent-browser network route "**/analytics.js" --abort
agent-browser network route "**/gtm.js" --abort
Resolve the annotation script path (checks common install locations):
ANNOT_JS="$(find . .claude .agents ~/skills -path '*/app-screenshots/references/annotate.js' 2>/dev/null | head -1)"
[ -z "$ANNOT_JS" ] && echo "ERROR: annotate.js not found — check skill installation"
For each page:
Step 1: Navigate
agent-browser open && agent-browser wait --load networkidle
For live sites, add agent-browser wait 2000 for lazy-loaded content. Chain when you don't need intermediate output: agent-browser open && agent-browser wait --load networkidle && agent-browser wait 2000
Step 2: Dismiss cookie/consent banners (live sites)
agent-browser find text "Accept" click
agent-browser wait "[class*=overlay]" --state hidden
For complex or iframed banners, see [references/live-sites.md](references/live-sites.md).
Step 3: Validate selectors BEFORE annotating
Always test selectors first. Failed selectors produce invisible annotations.
Use eval --stdin with heredoc for complex JS (avoids shell quoting issues):
agent-browser eval --stdin {
const el = document.querySelector(s);
if (!el) return 'MISSING: ' + s;
const r = el.getBoundingClientRect();
return 'OK: ' + s + ' (' + Math.round(r.width) + 'x' + Math.round(r.height) + ')';
}).join('\n')
EVALEOF
Quick match count check: agent-browser get count ""
If MISSING, use agent-browser snapshot -i -c to discover correct selectors and refs. Do not annotate with broken selectors. See [references/selectors.md](references/selectors.md) for tips.
For elements below the fold, scroll them into view first: agent-browser scrollintoview ""
Step 4: Inject annotations
Use eval --stdin to inject the annotation script safely:
agent-browser eval --stdin -.png
Zero-padded numbering: 01-homepage.png, 02-category.png, etc. Use --full for full-page screenshots when content extends below the fold.
Step 6: Verify screenshot quality
Read the screenshot file you just captured and check:
- Annotations are visible and not overlapping each other
- Labels are not cut off at viewport edges
- The page content loaded correctly (no spinners, broken images)
- Cookie banners or overlays are not blocking the view
If any issues: agent-browser eval "clearAnnotations()", fix the problem, re-annotate, and re-screenshot. Do not proceed with a bad screenshot.
Interactive screenshots
For search, menus, modals, hover states — use the snapshot → refs → interact pattern:
# 1. Snapshot to get refs
agent-browser snapshot -i
# 2. Interact using refs (more reliable than CSS selectors)
agent-browser click @e5 # Click search input by ref
agent-browser type @e5 "shoes"
agent-browser wait 2000
# 3. Annotate the result
agent-browser eval --stdin -screenshots.md` (live sites) or `docs/app-screenshots.md` (local).
Use numbered references **(1)**, **(2)**, **(3)** in descriptions to cross-reference the annotation badges visible in each screenshot.
```markdown
# - Visual Guide
Brief description of the site/app.
**URL:**
**Date:**
**Annotation key:** Red = primary actions, Blue = secondary elements, Green = content areas
---
##
1-3 sentences describing the page. Reference annotations by number:
"The **Search bar** (1) supports autocomplete. Use the **Category nav** (2) to browse departments.
The **Hero banner** (3) rotates seasonal promotions."
---
Phase 5: Compare environments (optional)
When documenting differences between staging and production, or before/after a redesign:
# Visual pixel diff between two URLs
agent-browser diff url https://staging.example.com https://prod.example.com
# Diff against a baseline screenshot
agent-browser diff screenshot --baseline docs/screenshots/01-homepage-before.png
# Accessibility tree diff (detect structural changes)
agent-browser diff snapshot
Include diff screenshots in the markdown to highlight what changed.
Phase 6: Cleanup
# Remove network intercepts
agent-browser network unroute
agent-browser close
Reference
- Selector tips and troubleshooting: See [references/selectors.md](references/selectors.md)
- Live site tips (cookies, lazy loading, rate limiting, geo): See [references/live-sites.md](references/live-sites.md)
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: alexanderop
- Source: alexanderop/app-screenshots
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.