agentleFS
lordmaculordmacu/nexo-rs/skills/browser/SKILL.md

Browser

Skill8 starsChanged 7 months agoApache-2.0View on GitHub

Drive a managed Chrome session via CDP. Navigate, click, fill forms, read the DOM, take screenshots, run JS.

68 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use V3 Core Implementation, V3 DDD Architecture and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ruvnetSkill

    V3 Core Implementation

    Core module implementation for claude-flow v3. Implements DDD domains, clean architecture patterns, dependency injection, and modular TypeScript codebase with comprehensive testing.

    Also in 15 projects that use this

  • ruvnetSkill

    V3 DDD Architecture

    Domain-Driven Design architecture for claude-flow v3. Implements modular, bounded context architecture with clean separation of concerns and microkernel pattern.

    Also in 15 projects that use this

  • +10 more from ruvnet

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 58 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use for tasks that require reading or interacting with a live web page: research, scraping structured data, filling forms, downloading …

    line 12

“Common” is measured: kept by at least 6 in 10 of the 58 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Browser

Controls a real Chrome/Chromium instance managed by the agent process. Use for tasks that require reading or interacting with a live web page: research, scraping structured data, filling forms, downloading files, grabbing a visual reference. Every tool call maps 1:1 to a Chrome DevTools Protocol command, so actions are immediate — no sandbox.

Use when

  • "Open example.com and tell me what the homepage says"
  • "Log in to this page and grab my account balance"
  • "Screenshot the dashboard for me"
  • "Find the cheapest flight on this comparison site"
  • "Click the Accept cookies button before you read the page"
  • "Fill this form with my saved details and submit"

Do not use when

  • You can answer with a simple HTTP GET — use fetch-url skill instead. The browser is heavy; spinning it up for plain JSON is wasteful.
  • Sites that ship a proper API (GitHub, Gmail, Slack). Use the matching skill (github, google, etc.) which talks the API directly.
  • Anti-bot protected sites (Cloudflare JS challenge, hCaptcha). The browser can technically get past some, but you're one bot-detection update from breaking. Ask the user first.

Tools exposed

ToolWhen to call
browser_navigate {url}Go to an absolute URL. Wait for load. Call first on any new page.
browser_snapshotText representation of the DOM with element refs (@e1, @e2…). Primary way to "see" a page. Refresh refs by calling again after any navigate/click.
browser_screenshotBase64 PNG of the viewport. For vision-capable models, or when layout matters. Don't call blindly — snapshot is cheaper.
browser_click {target}Click an element. target is an @eN from the latest snapshot or a CSS selector. Prefer @eN.
browser_fill {target, value}Replace the contents of an input / textarea. value is literal — special keys don't work here.
browser_press_key {key}Dispatch a keyboard event on the focused element. Names: Enter, Tab, Escape, Arrow{Up,Down,Left,Right}. Anything else is treated as a literal char.
browser_scroll_to {target}Scroll an element into view. Do this before clicking items the snapshot flagged as below the fold.
browser_wait_for {selector, timeout_ms?}Poll until a CSS selector appears. Use between browser_click and the next browser_snapshot when the page triggers an XHR/SPA transition.
browser_evaluate {script}Run arbitrary JS. Escape hatch for anything the other tools can't express. Returned value is in result.
browser_current_urlShortcut for browser_evaluate{ 'location.href' }.
browser_go_back / browser_go_forwardStep through the history stack.

Canonical workflow

1. browser_navigate({url: "https://example.com"})
2. browser_snapshot()                    ← read the page, collect refs
3. browser_fill({target: "@e12", value: "user@example.com"})
4. browser_fill({target: "@e13", value: "s3cret"})
5. browser_click({target: "@e14"})       ← submit
6. browser_wait_for({selector: ".dashboard"})
7. browser_snapshot()                    ← confirm logged in, collect new refs

Element refs (@eN) are only valid inside the snapshot that produced them — as soon as the DOM changes, re-call browser_snapshot before referencing more refs. CSS selectors stay valid across snapshots, but the LLM usually gets better accuracy with refs.

Failure modes

  • {ok: false, error: "navigate timed out"} — page didn't fire load within the timeout. Retry once with a longer implicit timeout via evaluate, or check if the URL is wrong.
  • {ok: false, error: "element not found"} — the @eN is stale or the selector doesn't match. Call browser_snapshot again.
  • {ok: false, error: "circuit breaker ... open"} — the plugin is cooling off after repeated failures. Wait a few seconds or ask the user to retry.

Gotchas

  • The browser is shared across agents that have plugins: [browser] — actions interleave. For side-effect-free reads it's fine; for stateful sessions (logins), give each agent its own user_data_dir in config/plugins/browser.yaml.
  • Screenshots are PNG base64, usually 100–500 KB. Don't pipe them through browser_evaluate — hit the tool directly.
  • browser_evaluate returns the value of the last expression. Wrap complex objects in JSON.stringify(...) if you need the full shape.
  • Headless mode is opt-in via browser.headless: true. With a visible window the user sees exactly what you're doing — useful for trust, expensive in CPU.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.