agent-browser
Use agent-browser for direct CLI browser automation, inspection, web-app testing, screenshots, sessions, and approved profile or CDP control. Trigger when a task names agent-browser or requires its CL
SKILL.mdskills/agent-browser/SKILL.md
GitHubagent-browser
The installed CLI is authoritative. Before browser work, run:
command -v agent-browser
agent-browser --version
agent-browser --help
agent-browser skills get core
If the CLI or requested skill is missing or unsupported, stop and report it. Do
not substitute stale instructions or invent flags. Load complete references only
when needed with agent-browser skills get core --full and
agent-browser skills path core.
Load a specialized installed skill when required, such as electron or
dogfood; fail closed if it is unavailable.
An explicit agent-browser request authorizes its managed Chromium runtime. Ask
for approval only for an explicit external executable, private profile, or CDP
endpoint. Use a task-unique session prefix, one stable named session per task,
and only owned-resource cleanup. With shared CDP, pin the session to its tab
using the upstream --pin-tab behavior.
Load agent-browser-policy with this skill. It supplements upstream core with
consent, privacy, profile, harness-versus-CLI, CDP, and cleanup rules.