Skip to content
Back to search

faster-chrome-devtools-skill

Agent skill that makes Chrome DevTools faster

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/zeke--faster-chrome-devtools-skill/badge.svg)](/listings/zeke--faster-chrome-devtools-skill)

HTML

<a href="/listings/zeke--faster-chrome-devtools-skill"><img src="/api/listings/zeke--faster-chrome-devtools-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Run this command to install the skill globally for all your installed agents:

npx skills add zeke/faster-chrome-devtools-skill --global --all --yes

This skill includes a dependency-free Node.js script that uses a WebSocket connection to Chrome, so **you will need Node.js installed**, but you **do not need** Chrome DevTools MCP, Puppeteer, or Playwright.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 114 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # faster-chrome-devtools-skill
  2. An [agent skill](https://agentskills.io) and command-line tool for controlling Chrome directly through the
  3. Chrome DevTools Protocol (CDP).
  4. Works with Claude Code, OpenCode, Codex, Pi, or any coding agent that supports the Agent Skills protocol.
  5. ![HIC LIMAX NAVIGAT LENTE](images/chrome-snail-09.jpg)
  6. ## Capabilities
  7. - List, open, and reuse tabs
  8. - Read compact accessibility snapshots with stable element references
  9. - Click and fill by accessibility reference or CSS selector
  10. - Navigate with explicit timeouts
  11. - Wait for text or selectors without arbitrary sleeps
  12. - Type into focused cross-origin frames using native CDP input
  13. - Capture compressed JPEG/WebP screenshots by default
  14. - Inspect console messages and failed network loads
  15. - Evaluate JavaScript or invoke any raw CDP method
  16. - Connect to authenticated remote browser endpoints
  17. - Hand Cloudflare Browser Run control to a human for
  18. [human-in-the-loop](https://developers.cloudflare.com/browser-run/features/human-in-the-loop/) sensitive actions.
  19. - Keep the connection alive in a lightweight background daemon
  20. ## Install
  21. Run this command to install the skill globally for all your installed agents:
  22. ```sh
  23. npx skills add zeke/faster-chrome-devtools-skill --global --all --yes
  24. ```
  25. This skill includes a dependency-free Node.js script that uses a WebSocket connection to Chrome, so **you will need Node.js installed**, but you **do not need** Chrome DevTools MCP, Puppeteer, or Playwright.
  26. ## Try it
  27. Once installed, you can invoke the skill by pasting one of these prompts into your coding agent.
  28. **🔑 Option 1: Drive your existing logged-in Chrome.**
  29. For this to work, you'll need to enable remote debugging in Chrome at `chrome://inspect/#remote-debugging`.
  30. ```text
  31. Using my logged-in Chrome, open https://github.com/notifications, snapshot the page, and summarize what needs my attention.
  32. ```
  33. **🕵️‍♀️ Option 2: Use a clean, anonymous local Chromium with no logins.**
  34. ```text
  35. Launch a fresh anonymous Chrome instance on a throwaway profile, open https://news.ycombinator.com, take a screenshot, and list the top five story titles.
  36. ```
  37. **⛅️ Option 3: Run in the cloud on Cloudflare [Browser Run](https://developers.cloudflare.com/browser-run/).**
  38. ```text
  39. Using Cloudflare Browser Run, open https://blog.cloudflare.com, read the page, and give me the five most recent post titles with their links as a markdown table. If the Browser Rendering credentials aren't set up yet, configure the required environment variables to authenticate first.
  40. ```
  41. The skill's Browser Run setup uses `lab=true` by default so Chrome beta features
  42. such as WebMCP are available when the remote browser supports them. It also
  43. supports structured Human in the Loop handoffs through Live View for login, MFA,
  44. CAPTCHA, sensitive input, and approval steps.
  45. ## Design
  46. The installable skill lives in `skills/faster-chrome-devtools-skill/`. The CLI is
  47. implemented entirely with Node.js built-ins. Its `scripts/lib/websocket.mjs`
  48. contains the small RFC 6455 client used to support custom HTTP upgrade headers,
  49. which Node's browser-compatible global `WebSocket` API does not expose.
  50. A loopback-only background daemon holds the CDP connection open for 20 minutes.
  51. Its random authentication token and connection details are stored in an
  52. owner-readable temporary state file. This avoids repeated Chrome access prompts
  53. without exposing the daemon on the network.
  54. Stop the sole active daemon with
  55. `node skills/faster-chrome-devtools-skill/scripts/cdp.mjs stop`. If several are
  56. running, the CLI lists safe daemon IDs and requires `stop --id <id>` or the
  57. explicit `stop --all`. An endpoint-specific stop can be selected with
  58. `--ws-endpoint` or `--http-endpoint`; cleanup never needs to reconnect to or
  59. rediscover the browser.
  60. ## Why a CLI instead of MCP?
  61. Google's [Chrome DevTools MCP](https://github.com/ChromeDevTools/chrome-devtools-mcp)
  62. is still great and this skill began as a guide for using it, but using a CLI
  63. instead has some key advantages:
  64. - Works anywhere. Any agent that supports Agent Skills and can run a shell can use
  65. it, with no MCP server to configure.
  66. - Full CDP protocol. MCP offers a fixed set of high-level tools plus JavaScript
  67. evaluation; the CLI's `raw` command can call any Chrome DevTools method.
  68. - Leaner context. The MCP server advertises dozens of tools to the model for the
  69. whole session. This skill loads only its description until used, then runs
  70. through the shell tool your agent already has.
  71. - No screenshot bloat. MCP returns screenshots inline to the model, which is why
  72. large ones can be silently dropped or even break a session. The CLI writes
  73. images to a file and returns the path, so the bytes enter the model only if you
  74. ask for them.
  75. To set up MCP instead, see [Browsers in the cloud](https://zeke.sikelianos.com/browsers-in-the-cloud/).
  76. ## Development
  77. See [AGENTS.md](AGENTS.md)
  78. ## More Skills
  79. Browse more skills at [zeke/skills](https://github.com/zeke/skills).
  80. ## License
  81. MIT

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.