Browserbase Browse.sh (Open Web Catalog)
The largest open-source catalog of skills for reliably performing tasks on real websites, plus a browser CLI for AI agents. Free. Source: Browserbase, browse.sh, 2026-05-18
Browserbase researched hundreds of sites and packaged each as a reusable "skill" that gives an agent the exact playbook to navigate that site: suggested DOM selectors and XHR/backend requests that, per Browserbase, cut token costs by 50x versus blind navigation. browse.sh is positioned as "the yellow pages of the internet" for agents. Source: Browserbase thread, 2026-05-18
How It Works
A single CLI handles skills, browser primitives, debugging, and cloud sessions:
browse skills add alltrails.com
browse skills add recreation.gov
browse skills add ramp.com
claude "Plan a road trip to Utah with EV charging stops and campsites, book and reimburse on Ramp."
Each catalog skill is tagged by execution kind: API (hit the site's JSON backend), Browser (drive the rendered page), Hybrid, or Fetch. Skills return structured JSON, for example AllTrails trail search, Amazon product search, Airbnb listing search via the SSR GraphQL blob, or NASA APOD. Browserbase ships verified skills built with partners (Ramp, Lovable, Interaction, Reducto), and the catalog is 100% free with open contribution. Source: Browserbase, browse.sh, 2026-05-18
The reviewed launch video shows the product shape more concretely than the tweet: browse.sh presents itself as "a browser CLI for your AI Agents", with tabs for web skills, browser primitives, debugging, and cloud sessions. The demo installs site skills such as apartments.com and craigslist.org, then routes an agent prompt through those installed playbooks instead of asking the model to rediscover the web page from pixels. Source: local video artifact review, 2026-07-01
Where It Fits Kevin's Stack
browse.sh is a skill library that pairs with whatever browser engine the agent drives. It complements Browser Testing Skills (Kevin's primary CDP automation) by supplying per-site playbooks instead of rediscovering selectors each run, the same amortization logic as Autobrowse (Browserbase Skills)'s persisted Browserbase playbooks. For change-watching rather than task execution, see Firecrawl Monitoring. The 50x token-cost claim is the same selector-over-vision economy Kevin already favors for agent web work.
Candidate Skill Routes
browser-trace and Browser to API are the high-leverage Browserbase skills from the 2026-06-30 bookmark audit. Together they turn a browser automation run into CDP/network evidence, then into an OpenAPI 3.1 hypothesis, HTML report, and client. Keep them candidate-gated rather than globally installed until redaction and skill-auditor checks pass, because Browserbase's own hardening commit notes that trace artifacts can contain cookies, auth headers, bearer tokens, passwords, private URLs, and authenticated screenshots. Source: GitHub browserbase/skills, commit d919e311d6ce8d8b3daa5f33853e67e43a545b9a, 2026-06-30
Use this route when the task is API discovery from an observed website flow. Use ordinary Browserbase site skills when the task is repeatable website operation. Use Browser Testing Skills when the task should become a deterministic regression.
Competitor Analysis Skill
The /competitor-analysis skill is a GTM-focused Browserbase workflow: discover competitors with Browserbase Search API, run a 4-lane Plan → Research → Synthesize pattern, and produce an HTML report with overview, per-competitor deep dives, a feature/pricing matrix, and a chronological mentions feed. It requires BROWSERBASE_API_KEY and the browse CLI, and the first-run permissions list is broad enough that this should remain an explicit opt-in skill rather than an always-loaded research route. Source: skills.sh Browserbase competitor-analysis, 2026-06-30
Current version/status snapshot: browse@0.9.1 latest on npm; browse@0.9.1-alpha-c18ab34 is the alpha tag as of 2026-07-01. The CLI package points at browserbase/stagehand packages/cli; default-branch HEAD is c18ab345e00451237ebd2bcf3567a76bbf1e52a3, and the browse@0.9.1 Git tag is 7224376585d6b2cc52e7a76b7035a7a5e0779320. The catalog skills repo remains browserbase/skills HEAD d919e311d6ce8d8b3daa5f33853e67e43a545b9a. Source: npm registry; git ls-remote, 2026-07-01
Site-Specific Skill Example: NerdWallet
The NerdWallet credit-card comparison skill shows the concrete product shape: a site-specific SKILL.md can accept a category, card name, URL, or free-form criteria and return structured card data including rating, fees, intro APR, welcome bonus, rewards, pros/cons, key benefits, canonical review URL, and captured affiliate Apply Now URL. The skill explicitly stays read-only and does not click Apply Now, sign in, or submit an application. Source: browse.sh skill page, 2026-06-30
The local video review matters because it shows browse.sh as an agent-facing catalog, not just a marketing page: the demo opens the Browserbase/browse.sh skill page, installs or references the NerdWallet workflow, then uses Claude Code to compare cards from structured site extraction. Keep this as a candidate route for repeatable website operations; do not import the NerdWallet skill globally because it is domain-specific, third-party maintained, and finance-adjacent. Source: X/@KenzieMac_Dev visual artifact, 2026-06-05
Timeline
- 2026-07-01 | Deep-reviewed the original browse.sh launch artifact, repaired the local thumbnail/manifest, reviewed the video demo, and refreshed source status. Decision: browse.sh remains a high-leverage web-skill catalog and browser CLI route; import individual skills only when a project needs that site workflow, with redaction/security review for any authenticated flow. Source: X/@browserbase, 2026-05-18; Source: npm/GitHub, 2026-07-01
- 2026-06-30 | Deep-reviewed the
/competitor-analysisbookmark, repaired the local thumbnail artifact, and added it as a candidate GTM/browser-research route. Decision: useful for blocked GTM/comparison work, not globally installed, because it needs Browserbase credentials and broad first-run CLI permissions. Source: X/@JaySahnan, 2026-06-18; Source: skills.sh Browserbase competitor-analysis, 2026-06-30 - 2026-06-30 | Deep-reviewed the NerdWallet compare-credit-cards browse.sh demo. Promoted the durable pattern: site-specific skills should declare read-only boundaries and return structured data; active installation remains candidate-gated. Source: X/@KenzieMac_Dev, 2026-06-05; Source: browse.sh skill page, 2026-06-30
- 2026-05-18 | Browserbase launched browse.sh, the largest open-source catalog of web-task skills, with a browser CLI for agents and verified partner skills (Ramp, Lovable, Interaction, Reducto). 2,020 likes, 3,370 bookmarks. Source: X/@browserbase, 2026-05-18
- 2026-06-30 | Added
browser-trace+ Browser to API as the API-discovery candidate route after reviewing the OpenTable demo, source snapshot, and trace-artifact security caveats. Source: X/@derekmeegan, 2026-05-13