ryanthedev/oberskills · Archived

browser

>- Controls a live Chrome browser through a persistent puppeteer-core MCP connection — click, type, hover, drag, fill forms; navigate pages; take screenshots and export PDFs; read the DOM and accessibility tree; run Lighthouse audits and Core Web Vitals traces; intercept, record, stub, or block network requests; export HAR traffic; emulate devices and network conditions; manage browser storage; upload and download files. Use when interacting with or automating a running browser, clicking elemen…

First seen Jul 16, 2026

Installation

$ npx skills add ryanthedev/oberskills --skill browser

Stronger alternatives

This repository is archived — consider an actively maintained alternative.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from ryanthedev/oberskills.

npx skills add ryanthedev/oberskills

Browse all from ryanthedev/oberskills

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Stars 60
License MIT
Default branch main
Open issues 0
Status Archived

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 5,364 B
  • docs SUMMARY.md 986 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 4 installs

SKILL.md

browser

Persistent Chrome/CDP control via puppeteer-core. The primary interaction model is snapshot → stable ref → action. Selectors and coordinates are fallbacks only. Large outputs (screenshots, DOM, AX trees, HAR, traces) are always written to disk — the tool returns a file path, never raw bytes.

Prerequisites

Call browserconnect before any other tool. Mode launch spawns (or reuses) a Chrome process; mode attach connects to a running Chrome via browserurl or ws_endpoint.

Primary loop: snapshot → ref → act

browser_snapshot          → returns compact AX tree; every interactive node carries a stable ref id
browser_click(ref=...)    → act on the ref (also: browser_type, browser_hover, browser_select, browser_fill_form, browser_press_key, browser_drag)
browser_snapshot          → verify the new state

A stale_ref error means the page changed; take a new snapshot. Selector/coordinate targeting is the fallback when a ref is unavailable.

Screenshot, DOM, and AX — route to a subagent when available

Screenshots, DOM trees, and accessibility trees are large (50 KB–5 MB). Loading them directly bloats the main context. Use the host's subagent tool when available: dispatch a small reviewer that reads the file path and returns a text summary, so the artifact never enters this conversation. If the host has no subagent surface, summarize inline only when the artifact is small enough; otherwise use local deterministic tools or ask how to proceed.

Dispatch Agent:
  model: haiku
  description: "browser: analyze [screenshot|dom|accessibility]"
  prompt: |
    The browser tool wrote an artifact to: <PATH>
    Read that file with the Read tool.
    Return a concise text summary: what the page shows, key elements,
    errors or state, and the specific answer to: <USER_QUESTION>
    Return text only — no raw HTML, JSON, or image data.

Tools that produce large artifacts: browserscreenshot, browserdom, browseraccessibility, browserexporthar, browserperformancestoptrace, browser_pdf. Pass the returned path to a subagent for reading when available.

Navigation and waiting

browsernavigate accepts http/https only. After navigation, use browserwait (strategy navigation or selector) to confirm the page settled before interacting. For text-triggered flows use browserwaitfor_text.

Tool surface by group

Full parameter details in the references below. Load the relevant one when planning work in that group.

Group Reference
Snapshot + refs interaction · Navigation · Read + extract references/interaction.md in this skill directory
Performance · Lighthouse · HAR · Network routing · Throttling references/perf-network.md in this skill directory
Storage · Device emulation · Geolocation · Capture (PDF / screencast / upload / download) references/storage-capture.md in this skill directory

Payload discipline

Artifact Where it lives In main context?
Screenshot PNG file path from browser_screenshot Prefer subagent; inline only if small enough
DOM HTML file path from browser_dom Prefer subagent; inline only if small enough
AX tree JSON file path from browser_accessibility Prefer subagent; inline only if small enough
HAR file file path from browserexporthar Prefer subagent; inline only if small enough
Performance trace file path from browserperformancestop_trace No — analyze via browseranalyzeinsight
PDF file path from browser_pdf Prefer subagent; inline only if small enough
Subagent text summary returned text Yes
Direct tool output (small) inlined in result Yes