stealthy-auto-browse

No file or documentation changes detected in this version.

**Input guidance and workflow recommendations improved for input modes.** - Now recommends using Playwright's `click` (CSS selector) first for most clicks—faster and more reliable. - Clearly explains that `system_click` should only be used when sites detect/block DOM events and that it requires calibration for coordinates. - Updated advice on when to use `fill` vs. `system_type` for form input. - Typical workflow updated to prioritize selector-based actions, with explicit fallback to system-level input only when strictly necessary. - No breaking changes to API or features.

- Added a new script: scripts/websearch.py. - Introduces new functionality related to web searching.

Version 1.7.0 - Added support for API authentication via `Authorization: Bearer <key>` header or `auth_token` query parameter if `AUTH_TOKEN` is set on the server. - Documented automatic request serialization in single-instance mode (requests are queued and run one at a time). - No file changes detected; changes are documentation improvements and clarifications in SKILL.md.

stealthy-auto-browse 1.6.0 introduces console log capture capabilities. - Added actions to enable, get, clear, and disable browser console logs, allowing retrieval of console messages (`console.log`, `console.error`, `console.warn`, etc.) with details including type, text, location, and timestamp. - New actions documented: `enable_console_log`, `get_console_log`, `clear_console_log`, `getclear_console_log`, `disable_console_log`. - No breaking changes to existing features.

- Added setup instructions and configuration details in a new [references/setup.md](references/setup.md) file. - Updated skill documentation for clarity, brevity, and organization. - Streamlined usage examples and input mode explanations. - Clarified when to use system (undetectable) vs. Playwright (detectable) input actions. - Consolidated API action references and improved quick-start workflow guidance.

add script mode

better wording

Rewrite SKILL.md as a proper usage guide instead of a dry API reference - Every action now explains what it actually does, not just a terse label - Documented response data fields for all actions so agents know what comes back - Explained the difference between system_click, mouse_click, and click (three ways to click, different tradeoffs) - Added parameter details (required/optional, defaults, valid values) for every action - Included example response JSON for complex returns (get_interactive_elements, get_network_log, get_cookies, list_tabs, get_last_dialog, get_last_download) - Added "When to use which" decision guide for system vs Playwright input - Documented gotchas: handle_dialog timing, calibrate after fullscreen, upload_file still needs form submit, storage is per-origin - Added full workflow section showing the typical interaction sequence - Documented container options, pre-installed extensions, page loaders with match rule logic

New actions: - Tabs: list_tabs, new_tab, switch_tab, close_tab - Dialogs: handle_dialog, get_last_dialog (alert/confirm/prompt) - Cookies: get_cookies, set_cookie, delete_cookies - Storage: get_storage, set_storage, clear_storage (local + session) - Downloads: get_last_download - Uploads: upload_file (Playwright set_input_files) - Network: enable/disable/get/clear_network_log - Waits: wait_for_element, wait_for_text, wait_for_url, wait_for_network_idle - Proxy: PROXY_URL environment variable for HTTP proxy support - XPath selector support (xpath= prefix) on all element actions

**stealthy-auto-browse 1.1.0 — Improved documentation, new usage guidance, and enhanced setup instructions** - Rewrote and reorganized documentation for clarity, including sections for when and when not to use this skill. - Added a detailed comparison with standard browser automation to highlight stealth advantages (no CDP exposure, OS-level input). - Clarified the setup process and included OpenClaw config and environment variable examples. - Expanded API examples with precise JSON payloads and usage tips for undetectable interaction. - Added container run examples for custom resolutions, timezones, persistent profiles, and URL auto-start. - Revised tips and best practices for maximizing stealth when automating sites with anti-bot protections.

remove runtime resulution setters

Initial release of stealthy-auto-browse: control a stealthy browser that evades bot detection using OS-level mouse/keyboard input. - Control a headless Firefox browser (Camoufox) that avoids common bot detection methods. - Supports both "system" (OS-level, undetectable) and Playwright (detectable) input modes. - Full API for navigation, mouse, keyboard, scrolling, screenshots, and state queries. - Includes workflow for undetectable interaction: navigate, discover elements and coordinates, perform mouse/keyboard input at the OS level. - Requires running the `stealthy-auto-browse` Docker container and setting `STEALTHY_AUTO_BROWSE_URL`.