browser-capture

Browser Resource Capture

You have access to a Playwright+CDP browser capture engine that archives every resource the browser receives during navigation. This is the CLI equivalent of Chrome's "Save All Resources" DevTools extension.

Preamble (run ONCE when skill is invoked)

_UPDATE=$(node "${CLAUDE_PLUGIN_ROOT}/scripts/update-check.js" 2>/dev/null || true)
[ -n "$_UPDATE" ] && echo "$_UPDATE" || true

If output shows UPDATE_AVAILABLE <old> <new>: tell the user "browser-capture v{new} is available (you're on v{old}). Run /plugin update browser-capture to upgrade." Then continue with the capture — don't block on updates.

If no output: you're up to date.

Setup (run ONCE before first capture)

Check if the capture engine is built. If not, build it:

node "${CLAUDE_PLUGIN_ROOT}/scripts/build.js" 2>&1

If output ends with "Ready to capture" or "Engine up to date": proceed.

If build fails: tell the user what went wrong and suggest running /browser-capture:health to diagnose.

This only runs once per session. After the engine is built, all subsequent captures use the existing bundle.

When to use this

The user wants to capture/record/archive what a browser loads. They care about:

• Every resource the browser receives (JS, CSS, HTML, fonts, images, API responses, XHR)
• Resources served from cache that never hit the network
• Full request/response headers and metadata
• The actual file contents, organized by domain
• Inline data URIs extracted from CSS/HTML

They do NOT care about:

• What the page "says" or its text content
• Summarizing or interpreting the page
• Completing tasks on the page
• Scraping structured data

How to use it

Run the capture engine. The bundled script self-resolves all paths:

NODE_PATH="${CLAUDE_PLUGIN_DATA}/node_modules" \
PLAYWRIGHT_BROWSERS_PATH="${CLAUDE_PLUGIN_DATA}" \
node "${CLAUDE_PLUGIN_DATA}/dist/capture.js" \
  --urls <url1> <url2> ... \
  --output ./browser-capture-output

If the bundle doesn't exist yet, fall back to the source script:

CLAUDE_PLUGIN_DATA="${CLAUDE_PLUGIN_DATA}" \
node "${CLAUDE_PLUGIN_ROOT}/scripts/capture.js" \
  --urls <url1> <url2> ... \
  --output ./browser-capture-output

To capture authenticated pages using cookies from the user's real browser:

NODE_PATH="${CLAUDE_PLUGIN_DATA}/node_modules" \
PLAYWRIGHT_BROWSERS_PATH="${CLAUDE_PLUGIN_DATA}" \
node "${CLAUDE_PLUGIN_DATA}/dist/capture.js" \
  --urls <url1> <url2> ... \
  --cookies-from chrome \
  --output ./browser-capture-output

Cookie flags:

• --cookies-from <browser> — import cookies from a real browser (chrome, brave, edge, chromium, arc, vivaldi, opera)
• --cookie-domains <d1,d2> — only import cookies for specific domains (comma-separated)
• --cookie-profile <name> — browser profile (default: Default)

The output is a domain-organized directory tree. Every resource gets a real file at its URL path. Resources where content couldn't be retrieved get a No Content: {url} marker file. A _metadata.json sidecar has the full request/response headers, status, timing, resource type, and cache status for every entry.

Critical behavior

1. The archive IS the deliverable. Do not summarize, analyze, or interpret the captured content unless the user asks you to after the capture is complete.
2. Capture everything by default. Beautify, all domains, no-content markers, data URI extraction — all ON unless the user says otherwise.
3. Your job during capture is navigation, not task completion. The browser is an instrument for recording, not a tool for accomplishing tasks on the site. Navigate to the pages the user described so the capture has what it needs.
4. After capture, report what was captured: total files, size, domains hit, resource type breakdown. Then ask if the user wants to filter, grep, or analyze anything from the capture.
5. Cookie import is opt-in. If the user mentions needing to be logged in, authenticated, or access private pages, use --cookies-from <browser>. Ask which browser if unclear. Never import cookies unless the user indicates they need authentication.
6. Cookie import is non-fatal. If cookie import fails (browser not found, DB locked, decryption error), the capture continues without cookies. Report the failure but don't abort.