Vibeleaderboard
Index / app
Visit github.com
Category
AI Agents
Rank
Type
APP
Date

About

A set of Claude Code skills for browser automation through Browserbase and the official bb CLI — driving remote browser sessions with anti-bot stealth, CAPTCHA solving, and residential proxies.

What it does

Skills is a broad automation toolkit for coding agents. It can inspect and operate websites, capture browser internals, derive API descriptions from observed traffic, generate repeatable scripts, test interfaces, and run structured company or competitor research. Individual skills combine agent instructions with command-line scripts and staged artifact pipelines.

Why it's ranked here

The strongest case is breadth backed by concrete implementation. Traces feed API discovery and iterative script generation, while separate skills cover testing, migration, authenticated sessions, and research. The repository also includes automated structural validation. The main reservation is operational weight: advanced workflows depend on external services, credentials, and generated evidence that remains observational rather than authoritative.

What's good

The trace tooling preserves network, console, runtime, page, DOM, and target events in searchable page buckets. API discovery filters noise, joins request and response bodies, infers schemas, redacts examples, records confidence, and emits OpenAPI 3.1 plus reports and a client. Autobrowse caches generation inputs, verifies scripts in fresh sessions, and can retry after failures.

Tradeoffs

Remote runs and verification require Browserbase credentials, while iterative evaluation and code generation require Anthropic credentials. Local operation requires Chrome. Captured traces may contain cookies, authorization headers, passwords, private URLs, and authenticated screenshots, although the evaluation harness restricts their filesystem permissions. Generated API descriptions are best-effort observations, and response schemas are unavailable when response bodies were not captured.

How to use it well

It suits engineers who want an agent to investigate browser behavior, reproduce navigation, audit changed interfaces, or bootstrap an API client from real traffic. Start with a trace, query failures and hosts, then feed the evidence into testing, API discovery, or script generation. Treat inferred specifications as leads, not contractual documentation.

Technical notes+

README.md lists 16 skills and documents installation through both npx skills add browserbase/skills and the Claude Code plugin marketplace. package.json declares an ES module package, a CLI binary, a TypeScript build script, and a postinstall build. scripts/validate-skills.mjs is a zero-dependency Node 20+ validator that checks frontmatter, per-skill MIT licensing, banned entries, documentation links, and size warnings. skills/browser-trace/scripts/lib.mjs defines shared JSONL handling and CDP event buckets, while skills/browser-trace/scripts/query.mjs exposes page, error, host, and timeline queries. skills/autobrowse/scripts/evaluate.mjs limits its agent tool to parsed browse commands executed without a shell and applies owner-only permissions to trace artifacts. skills/autobrowse/scripts/codegen.mjs builds cached Anthropic prompts from tasks and traces, emits framework scaffolds, and optionally verifies and rewrites generated scripts. skills/browser-to-api/scripts/discover.mjs dispatches the load, filter, normalize, infer, and emit pipeline; skills/browser-to-api/scripts/emit.mjs produces OpenAPI 3.1 documents, confidence metadata, reports, HTML, and a generated client.

Observed

License policy
Every skill is validated for MIT frontmatter and an MIT license text naming Browserbase, Inc.
Primary implementation language
Repository automation shown here uses JavaScript ES modules, with TypeScript compilation configured for the package.
Install surface
Installable through npx skills and through the Claude Code plugin marketplace.
Interfaces
Provides Claude Code skills and command-line workflows centered on the browse CLI.
Platform support
Chrome setup instructions cover macOS, Windows, and Linux.
Skill collection
The documented plugin contains 16 skills spanning browsing, tracing, testing, API discovery, code generation, migration, and research.

Read from README.md, package.json, scripts/validate-skills.mjs, skills/browser-trace/scripts/lib.mjs, skills/autobrowse/scripts/codegen.mjs, skills/autobrowse/scripts/evaluate.mjs, skills/browser-to-api/scripts/emit.mjs, skills/browser-to-api/scripts/load.mjs, skills/browser-trace/scripts/query.mjs, skills/browser-to-api/scripts/infer.mjs, skills/browser-to-api/scripts/filter.mjs, skills/autobrowse/scripts/unify-trace.mjs, skills/browser-to-api/scripts/discover.mjs.

What it can do

  • Automate web browser interactions with anti-bot stealth and CAPTCHA solving

    CLI commands and target website URLsAutomated browser actions on remote Browserbase sessions

  • Deploy serverless browser automation to cloud

    Browser automation scriptsDeployed serverless functions on Browserbase cloud

  • Diagnose and fix failing browser automations

    Failing browser automation scriptsSite-specific playbook with fixes for bot detection, selectors, timing, auth, and captchas

  • Capture full browser automation traces with DevTools protocol

    Browser automation sessionCDP traces, screenshots, DOM dumps organized in searchable per-page buckets

  • Sync cookies from local Chrome to persistent browser context

    Local Chrome browser cookiesAuthenticated Browserbase session with synced cookies

  • Fetch HTML or JSON from web pages without browser session

    Website URLsHTML/JSON content with status codes, headers, and redirect information

  • Search the web and return structured results

    Search queriesStructured search results with titles, URLs, and metadata

  • Perform AI-powered adversarial UI testing on web applications

    Git diffs or web application URLsBug reports and test results from automated UI exploration

Intel on Skills

More in Intel

Tech Stack

Node.jsTypeScript

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.