-
Notifications
You must be signed in to change notification settings - Fork 1.6k
Description
Environment
- OS: Ubuntu 24.04.4 LTS (Server/NAS environment)
- agent-browser version: 0.22.3 (npm installed globally)
- Chrome version: 147.0.7727.24 (installed via
agent-browser install) - Node.js: v22
- Running as: OpenClaw AI agent (https://github.com/openclaw/openclaw)
- Installation path: ~/.npm-global/lib/node_modules/agent-browser/bin/agent-browser-linux-x64
System Dependencies
All system dependencies are already installed:
libxcb-shm0, libx11-xcb1, libx11-6, libxcb1, libxext6, libxrandr2,
libxcomposite1, libxcursor1, libxdamage1, libxfixes3, libxi6,
libgtk-3-0t64, libpangocairo-1.0-0, libpango-1.0-0, libatk1.0-0t64,
libcairo-gobject2, libcairo2, libgdk-pixbuf-2.0-0, libxrender1,
libasound2t64, libfreetype6, libfontconfig1, libdbus-1-3, libnss3,
libnspr4, libatk-bridge2.0-0t64, libdrm2, libxkbcommon0, libatspi2.0-0t64,
libcups2t64, libxshmfence1, libgbm1, fonts-noto-color-emoji,
fonts-noto-cjk, fonts-freefont-ttf
Bug Description
Running agent-browser open <url> hangs indefinitely with no output. This is observed when running inside OpenClaw AI agent on a Linux Server/NAS environment.
Steps to reproduce:
$ agent-browser open example.com
# hangs forever, no outputExpected behavior:
- Chrome should launch in headless mode
- CLI should return and allow subsequent commands
Actual behavior:
- CLI hangs indefinitely (tested up to 60s timeout)
- No error message is shown initially
- The daemon never starts (no socket file created at
~/.agent-browser/socket*)
Error message (after ~30s of retries):
Failed to connect: No such file or directory (os error 2) (after 5 retries - daemon may be busy or unresponsive)
This error appears after the CLI finally gives up, but only when running in background mode. In foreground mode, the command simply hangs forever with no output.
Analysis
- Chrome binary is present and executable: ✓
- No socket file is ever created:
find ~/.agent-browser -name "*.sock"returns nothing - PR fix(unix): detect stale unix socket by attempting connection #114 is already included in 0.22.3 — the fix adds a 50ms timeout on socket connection check, but doesn't solve the root cause (daemon never starts)
- Chrome process doesn't even start — verified via
ps aux | grep chromeshowing 0 processes during the hang - This may be related to:
- Issue Daemon crash causes orphaned processes and connection failures #113 (daemon crash causes orphaned processes) — though no orphaned processes are left
- Running in a headless server/NAS environment without a display
Additional Context
agent-browser --versionworks fineagent-browser installandagent-browser install --with-depscomplete successfully- The daemon fails silently before attempting to launch Chrome
- This environment is a NAS (possibly running in a container or minimal server environment)
- This is being reported from an OpenClaw AI agent running on this Linux Server/NAS
Possible Root Cause Hypothesis
The daemon process may be failing to start entirely in this environment, possibly due to:
- Missing display/wayland environment variables
- Chrome needing
--no-sandboxflag in this environment but daemon not passing it - Some other Linux server/NAS-specific restriction
Would like to help with
I'm happy to run additional diagnostics, test patches, or provide more information if needed.