Drive a Real Browser with opencli-browser

Where this fits

You're building an agent that must verify a logged-in dashboard, fill a multi-step form, or extract data from a site that blocks headless browsers. Your agent can't rely on APIs or static HTML—it needs a real Chrome session. The opencli-browser skill gives you a CLI-driven browser that returns structured, agent-readable envelopes instead of raw DOM dumps. It's the missing glue between an agent's intent and a live, interactive page.

Why agents benefit

• Selector-first contract with confidence scores: Every command returns a match confidence level, so the agent knows when to retry or ask for clarification instead of guessing.
• Compound form fields handled: Handles date pickers, dropdowns, and multi-part inputs as single logical fields, reducing brittle step sequences.
• Stale-element handling: Automatically re-queries the DOM when a reference goes stale, so the agent doesn't crash mid-flow.
• Network capture: Intercepts requests and responses, letting the agent verify API calls or extract data from XHR payloads.
• Agent-native envelopes: Each response includes status, matched, suggested_next—explicit guidance that keeps the agent on track.

Practical scenarios

Verify a post-login redirect

After submitting credentials, the agent needs to confirm the redirect landed on the expected dashboard and that the user's name appears. Using opencli browser navigate and opencli browser query with a selector, the envelope tells you if the element matched and how confident it is.

Fill a multi-step checkout form

A checkout with shipping, payment, and review steps can be driven sequentially. The skill's compound field handling lets the agent set a date or select a country without splitting it into multiple low-level actions.

Extract paginated table data

For a site with paginated results, the agent can loop through pages, capture the table rows, and use the network capture to also pull the underlying JSON API responses—all in one session.

Add it to your agent workflow

Install via the OpenAgentSkill CLI:

npx skills add jackwener/OpenCLI --skill opencli-browser

Then run opencli doctor to verify the environment. A basic interaction looks like:

opencli browser navigate --url https://example.com/login
opencli browser fill --selector "#email" --value "agent@example.com"
opencli browser click --selector "button[type=submit]"
opencli browser query --selector ".dashboard-title"

The envelope returned from query tells you if the title matched and how confident it is—no guesswork. For more context on how this skill fits into the broader agent ecosystem, see the skill page on OpenAgentSkill.

Compare before adopting

Before committing, compare opencli-browser with alternatives like Playwright MCP or Puppeteer-based tools. Look at the quality of the envelope format—does it give actionable next steps or just raw selectors? Check maintenance freshness: the repo has 28k stars and recent activity, but verify the last commit. Also assess workflow fit: if you already have a browser automation stack, does this slot in or require a rewrite?

Why it is worth tracking

With 28k stars and an active author, opencli-browser is gaining traction as a practical, agent-first browser driver. Its design philosophy—envelopes over raw output—aligns with how modern agents need to reason about actions. Even if you don't adopt it today, it's worth evaluating when you next need live-browser automation that your agent can trust. The skill's explicit focus on agent-native contracts makes it a strong candidate for production workflows.