Web Agent

Browser automation agent package for IQuly using agent-browser.

What it does

• Opens and navigates websites
• Captures compact accessibility snapshots with deterministic refs
• Clicks, fills, types, presses keys, and waits for page state changes
• Uses semantic find commands when refs are unavailable
• Reads page text, attributes, visibility, URLs, and titles
• Saves screenshots and PDFs into the workspace
• Supports isolated browser sessions for separate tabs, auth state, or concurrent tasks
• Supports session state save/load, per-origin headers, device/theme emulation, and structural diffing

Tool model

• browser_run is the shaped browser automation surface
• fs_* tools are retained so the agent can manage saved screenshots, PDFs, and notes
• package_* tools are retained so the agent can inspect its packaged playbook files when needed
• Broad shell access is intentionally removed

Package layout

web-agent/
  AGENT.md
  manifest.json
  Dockerfile
  tools.json
  schedule.json
  workspace-config.json
  skills/
    browser-workflow/
      SKILL.md
  files/
    browser-playbook.md
    team-context.md
  tools/
    browser-run/
      run.js

Notes

• Runtime dependency installation is handled in the package Dockerfile.
• agent-browser runs with compact ref-based interaction patterns that are efficient for LLMs.
• The shaped wrapper supports the core navigation, interaction, wait, find, screenshot, tab, cookie, network request inspection, state persistence, variant emulation, and diff flows used most often by the agent.
• Use iquly push . --dry-run before publishing.