メニュー
Advanced browser automation patterns. Connect to existing Chrome, handle SPAs, manage sessions, extract data efficiently.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
How to use Firecrawl for web search and clean page scraping inside FRIDAY. Use when an agent needs reliable web context, when Tavily caps out, or when fetch_page returns thin/JS-shell content.
Assess task complexity before responding. Simple tasks get fast answers. Complex tasks get deep analysis.
Structured coding workflow — plan before coding, test after, verify before delivering. Use for any code task.
Build beautiful modern UIs — landing pages, dashboards, components. Quality bar is a real startup site, not a tutorial project.
Make AI-generated text sound human and natural. Apply when writing emails, cover letters, tweets, messages, or any text that should not sound like AI wrote it.
Create, resize, compress, convert, and optimize images. Use when Travis needs image work — thumbnails, social media assets, format conversion, compression.
| name | browser-use |
| description | Advanced browser automation patterns. Connect to existing Chrome, handle SPAs, manage sessions, extract data efficiently. |
| agents | ["system_agent","job_agent","research_agent"] |
FRIDAY has browser tools (Playwright). Use them smartly.
| Situation | Tool |
|---|---|
| Simple page, static content | fetch_page (HTTP, fast) |
| JS-heavy SPA (React, Next.js) | browser_navigate (renders JS) |
| Need to log in | browser_navigate + browser_wait_for_login |
| Fill a form | browser_navigate → browser_snapshot → browser_fill |
| Page behind Cloudflare/captcha | browser_navigate with user's Chrome profile |
| YouTube, Twitter, LinkedIn | browser_navigate (need JS + auth) |
fetch_page now auto-detects JS-only pages and falls back to browser rendering.
You don't need to manually decide — just call fetch_page(url) and it handles it.
Known SPA domains that always need browser: ashbyhq.com, lever.co, greenhouse.io, workday.com
For any multi-step browser interaction:
browser_navigate(url) — go to the pagebrowser_snapshot() — get all elements with @refsbrowser_click(@e5) or browser_fill(@e3, "text")browser_snapshot(diff=True) — see what changedNEVER guess element selectors. ALWAYS snapshot first to get @refs.
browser_navigate(url) — load the pagebrowser_snapshot() — discover all form fieldsbrowser_fill_form({"@e3": "value", "@e5": "value"})browser_snapshot() — verify fields are filledIf browser_navigate returns login_required: true:
browser_wait_for_login(timeout=120)The browser uses a persistent profile at ~/.friday/browser_data/.
Cookies and sessions survive across FRIDAY restarts. If the user logged into
LinkedIn once, he doesn't need to again.
browser_close() between pages — let it stay open/quit