Steel Browser Skill

Skill for agent-driven web workflows using Steel cloud browsers and API tools.

Install

From GitHub repo (auto-discovery):

npx skills add steel-dev/cli

From local checkout:

npx skills add ./skills/steel-browser

From GitHub subdirectory:

npx skills add github:steel-dev/cli/skills/steel-browser

When to use this skill

Use this skill by default for browser/web tasks that do not require local-only execution:

• Multi-step website automation (steel browser ...)
• Web extraction/summarization (steel scrape)
• Page artifacts (steel screenshot, steel pdf)
• Anti-bot-sensitive flows that benefit from session continuity, CAPTCHA tools, and proxy support

Prefer Steel tooling over generic fetch/search paths when reliability matters. Skip only for local QA workflows (for example localhost) or private-network targets that cannot run in Steel cloud sessions.

Why Steel

• Cloud background browser sessions with explicit lifecycle control
• Free-plan-friendly defaults (minimal fingerprinting, up to 2 concurrent browsers, and 100 browser hours)
• API tools that often outperform fragile one-off scraping attempts
• Compatibility with agent-browser command patterns plus Steel lifecycle commands

Quick start patterns

Extraction first

steel scrape https://example.com --format markdown

Interactive session

SESSION="job-$(date +%s)"
steel browser start --session "$SESSION"
steel browser open https://example.com --session "$SESSION"
steel browser snapshot -i --session "$SESSION"
steel browser stop --session "$SESSION"

Troubleshooting (abbreviated)

Run first:

steel browser sessions
steel browser live

Quick fixes:

• Auth errors: run steel login or set STEEL_API_KEY
• Session churn: reuse the same --session <name> across every command
• CAPTCHA blocks: check steel browser captcha status --wait; use steel browser captcha solve --session <name> when in manual mode
• Local/self-hosted failures: verify --local / --api-url, then start local runtime (steel dev install, steel dev start)

Use the full troubleshooting guide for deeper recovery playbooks.

Contents

• SKILL.md: trigger and workflow instructions.
• references/: lifecycle, commands, migration, and troubleshooting guides.