Gologin Agent Browser Skill
Use this skill when the task needs browser state, interaction, or multi-step navigation.
Core Rules
- Use this skill for browser automation, not for scraping-only tasks.
- Prefer snapshot refs like
@e3aftersnapshot. - Refresh the snapshot after navigation or DOM-changing actions.
- Use semantic
findwhen stale refs or dynamic pages make raw refs unreliable. - Keep browser session work inside the
gologin-agent-browserCLI model.
CLI Surface
Package:
gologin-agent-browser-cli
CLI:
gologin-agent-browser
Main command families:
- session lifecycle:
open,close,sessions,current - ref loop:
snapshot,click,type,fill - navigation helpers:
wait,scroll,scrollintoview - semantic helpers:
find,get - artifact helpers:
screenshot,pdf,upload
Setup
Expected environment variables:
GOLOGIN_TOKENGOLOGIN_PROFILE_IDoptionalGOLOGIN_DAEMON_PORToptionalGOLOGIN_CONNECT_BASEoptional
Tool Map
| Skill tool | CLI command | Use when |
|---|---|---|
agent_browser_open | gologin-agent-browser open <url> | A browser session must start |
agent_browser_snapshot | gologin-agent-browser snapshot | The next actionable refs are needed |
agent_browser_click | gologin-agent-browser click <target> | A target should be clicked |
agent_browser_type | gologin-agent-browser type <target> <text> | Text should be typed into a target |
agent_browser_fill | gologin-agent-browser fill <target> <text> | An input should be filled in one step |
agent_browser_find | gologin-agent-browser find ... | Semantic targeting is more reliable than refs |
agent_browser_get | gologin-agent-browser get <kind> [target] | Data should be extracted from the page |
agent_browser_wait | gologin-agent-browser wait ... | The flow should wait on state, text, or load |
agent_browser_upload | gologin-agent-browser upload <target> <file...> | Files should be uploaded |
agent_browser_pdf | gologin-agent-browser pdf <path> | A PDF artifact is needed |
agent_browser_screenshot | gologin-agent-browser screenshot <path> | A screenshot artifact is needed |
agent_browser_close | gologin-agent-browser close | The session should end |
agent_browser_sessions | gologin-agent-browser sessions | All active sessions should be inspected |
agent_browser_current | gologin-agent-browser current | The active session summary is needed |
Operating Pattern
Ref Loop
- Open a session with
agent_browser_open. - Capture refs with
agent_browser_snapshot. - Act with
agent_browser_click,agent_browser_type, oragent_browser_fill. - If the result says
snapshot=stale, refresh withagent_browser_snapshot. - Repeat until the task is complete.
Semantic Loop
- Use
agent_browser_findwhen the page is dynamic or refs are brittle. - Use
agent_browser_getwhen a value, title, HTML, or URL should be read back. - Use
agent_browser_waitwhen the flow must pause for text, load state, or target availability.
Artifact Loop
- Use
agent_browser_screenshotwhen a visual artifact is needed. - Use
agent_browser_pdfwhen a document artifact is needed. - Use
agent_browser_uploadwhen the site requires file input.
Snapshot Discipline
- Refs are best-effort and local to the latest snapshot.
- Do not reuse old refs after navigation or heavy DOM changes.
- Prefer a fresh snapshot or semantic
findif a ref stops resolving.
References
- See
tools.mdfor command contracts. - See
examples/for short usage examples. - See
workflows/for repeatable browser patterns.