use attribute selector instead of id selector to avoid css escaping issues with special chars (e.g. :r1:). parse aria-controls/aria-owns as space-separated list per aria spec.
Portal-based dropdowns render options in a separate container via aria-controls. Resolve aria-controls/aria-owns to find options outside the trigger element's descendants.
Use separate system prompts for Opus 4.5 and Haiku 4.5 that are at least
4096 tokens (which is the token minimum for both models' input caching),
enabling input caching and notably reducing cost.
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Enable input caching for Claude Opus 4.5 and Haiku 4.5 by switching to
dedicated 4096+ token system prompts. This lowers cost and improves
performance for these models.
- **New Features**
- Detect Opus/Haiku 4.5 variants and load long system prompts (standard,
flash, no-thinking).
- Add 4096+ token templates: system_prompt_anthropic.md,
system_prompt_anthropic_flash.md,
system_prompt_anthropic_no_thinking.md.
- Pass model_name to SystemPrompt to drive correct template selection.
<sup>Written for commit f85bbb49c8.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
Resolves [#3557](https://github.com/browser-use/browser-use/issues/3557)
and similar issues. extract_clean_markdown() does extract DOM elements
not visible; its extraction was just being overly truncated.
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Increased the extraction character limit from 30,000 to 100,000 to
prevent valid content from being cut off during markdown extraction.
Resolves#3557 by allowing full page content (including non-visible DOM
elements) to be returned without premature truncation.
<sup>Written for commit 80a08a4a03.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- CURSOR_SUMMARY -->
---
> [!NOTE]
> **Increase extraction capacity**
>
> - Bumps `MAX_CHAR_LIMIT` in `extract()` (in
`browser_use/tools/service.py`) from `30000` to `100000` to allow larger
page content to be processed without truncation.
>
> *Effect:* fewer cut-offs when extracting long pages; may increase
processing time/token usage.
>
> <sup>Written by [Cursor
Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit
80a08a4a03. This will update automatically
on new commits. Configure
[here](https://cursor.com/dashboard?tab=bugbot).</sup>
<!-- /CURSOR_SUMMARY -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Adds thinking_level support for Gemini 3 (Pro and Flash) with safe
defaults and warnings to prevent misconfiguration.
- **New Features**
- Adds thinking_level for Gemini 3; Pro supports low/high (default low),
Flash supports minimal/low/medium/high (default minimal).
- Uses thinking_level for Gemini 3 and thinking_budget for Gemini
2.5/flash.
- Warns on unsupported configs (budget on 3, level on 2.5); falls back
to low if minimal/medium set on 3 Pro.
- Sets thinking_budget=0 by default for Gemini 2.5/flash when not
provided.
- **Migration**
- For Gemini 3, use thinking_level (Pro: low/high; Flash:
minimal/low/medium/high); stop using thinking_budget.
- For Gemini 2.5/flash, use thinking_budget (0 disable, -1 dynamic, or
token count).
<sup>Written for commit b63ae37008.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
## Summary
- Add `ChatBrowserUse` as the first LLM option in the CLI `run` command
when `BROWSER_USE_API_KEY` is set
- Update error message to mention `BROWSER_USE_API_KEY` as a valid
option
## Problem
The CLI `run` command currently only checks for `OPENAI_API_KEY`,
`ANTHROPIC_API_KEY`, or `GOOGLE_API_KEY`, but the
[documentation](https://docs.browser-use.com/supported-models) shows
that `ChatBrowserUse` with `BROWSER_USE_API_KEY` is a valid and
recommended LLM option for browser automation.
Users who only have `BROWSER_USE_API_KEY` configured get the error:
```
No LLM configured. Set OPENAI_API_KEY, ANTHROPIC_API_KEY, or GOOGLE_API_KEY
```
## Solution
Added `ChatBrowserUse` support in `get_llm()` function, checking
`BROWSER_USE_API_KEY` first since it's the optimized model for
browser-use tasks.
## Testing
Tested locally - `browser-use run "go to google.com"` now works with
only `BROWSER_USE_API_KEY` set.
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Adds ChatBrowserUse support to the CLI run command by preferring
BROWSER_USE_API_KEY in get_llm, enabling runs with only that key.
Updates the “No LLM configured” error to include BROWSER_USE_API_KEY.
<sup>Written for commit 2774be5cc6.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Set the default thinking_budget to 0 for Gemini 3 models when not
specified, matching the existing behavior for flash/flash-lite. This
prevents unexpected thinking token usage and keeps responses fast.
<sup>Written for commit 5b06f4edd9.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Add support for the bu-2-0 Browser Use model so agents can use the
latest LLM.
- **New Features**
- Allow bu-2-0 in ChatBrowserUse model validation (replaces bu-1-5).
- Update examples/simple.py to use bu-2-0.
- **Migration**
- If you used bu-1-5, switch to bu-2-0.
<sup>Written for commit f3aad3c145.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Adds support for the new bu-1-5 model in ChatBrowserUse and updates the
simple example to use it. Existing models (bu-latest, bu-1-0) and the
browser-use/* pattern remain supported.
<sup>Written for commit 37586c9e9f.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
The CLI 'run' command now supports BROWSER_USE_API_KEY as an LLM option,
using ChatBrowserUse which is optimized for browser automation tasks.
This aligns the CLI behavior with the documented supported models at
https://docs.browser-use.com/supported-models which shows ChatBrowserUse
as a valid LLM option using BROWSER_USE_API_KEY.
Changes:
- Add ChatBrowserUse as first LLM option in get_llm() when BROWSER_USE_API_KEY is set
- Update error message to mention BROWSER_USE_API_KEY
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Adds a fast, persistent CLI for browser automation with a background
session server. Enables quick, multi-step workflows with commands for
navigation, interaction, tabs, screenshots, JavaScript, Python, and
agent tasks.
- **New Features**
- New "browser-use" CLI (aliases: bu, browser) with a session server
over Unix sockets/TCP for low-latency commands.
- Commands: open, state (LLM-style output), click, type, input, scroll,
back, screenshot, switch, close-tab, keys, select, eval, extract;
sessions, close (--all), and server status/stop/logs; JSON output
supported.
- Python execution with a persistent namespace and a browser helper
(goto, click, type, scroll, screenshot, html).
- Agent task runner (requires API key) with interactive API key prompt
and config storage.
- Browser modes: chromium (headed/headless), real Chrome with profiles,
and remote cloud; added CLI docs and Claude Code SKILL.md.
- **Migration**
- CLI entry points updated: use browser-use as the main command; legacy
TUI is now browser-use-tui.
- For remote mode and agent tasks, set BROWSER_USE_API_KEY or add
{"api_key": "..."} to ~/.config/browser-use/config.json.
<sup>Written for commit b6bdc17bcb.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
- State command outputs same format that browser-use agents see
- Added aliases: bu, browser (all work the same)
- Headed mode now correctly shows browser window
- Change main entry point to 'browser-use'
- Move TUI to 'browser-use-tui'
- Add README with human instructions
- Update SKILL.md references
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Update custom tool registry documentation to clarify correct parameter
naming and function signatures.
The previous documentation incorrectly showed `browser: Browser` as an
injectable parameter for custom tools. The agent injects parameters by
name matching, so the correct parameter name is `browser_session:
BrowserSession`. This PR fixes this, recommends async functions, and
adds a "Common Pitfalls" section to guide users.
---
[Slack
Thread](https://browser-use.slack.com/archives/C08TZGLPLER/p1768767006926899?thread_ts=1768767006.926899&cid=C08TZGLPLER)
<a
href="https://cursor.com/background-agent?bcId=bc-a9112a17-72ed-4ef2-acba-8d9873477a89"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/open-in-cursor-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://cursor.com/open-in-cursor-light.svg"><img alt="Open in
Cursor"
src="https://cursor.com/open-in-cursor.svg"></picture></a> <a
href="https://cursor.com/agents?id=bc-a9112a17-72ed-4ef2-acba-8d9873477a89"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://cursor.com/open-in-web-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://cursor.com/open-in-web-light.svg"><img alt="Open in Web"
src="https://cursor.com/open-in-web.svg"></picture></a>
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Clarified custom tool docs to use the correct `browser_session:
BrowserSession` injection and recommend async functions with
`ActionResult`, preventing silent tool failures. Added warnings and a
Common Pitfalls section to make name-based parameter injection clear.
- **Bug Fixes**
- Fixed examples to use `browser_session: BrowserSession` (not `browser:
Browser`), return `ActionResult`, and be `async`.
- Added warnings about name-based injection and a Common Pitfalls
section with wrong/correct examples.
- Updated AGENTS.md and tools docs (add.mdx, basics.mdx) for
consistency.
<sup>Written for commit 1c70a9dbc5.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- CURSOR_SUMMARY -->
---
> [!NOTE]
> Clarifies custom tool docs and examples to correctly use
`browser_session: BrowserSession`, `async` functions, and
`ActionResult`, reducing silent failures from name-based injection
mismatches.
>
> - Update examples in `AGENTS.md`, `docs/customize/tools/basics.mdx`,
and `docs/customize/tools/add.mdx` to replace `browser: Browser` with
`browser_session: BrowserSession`, switch to `async`, and return
`ActionResult`
> - Add prominent warnings about name-based parameter injection and a
new **Common Pitfalls** section with wrong/correct examples
> - Minor note updates to reference deterministic Actor actions via
`browser_session`
>
> <sup>Written by [Cursor
Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit
1c70a9dbc5. Configure
[here](https://cursor.com/dashboard?tab=bugbot).</sup>
<!-- /CURSOR_SUMMARY -->
- Fix basics.mdx: change browser: Browser to browser_session: BrowserSession
- Fix basics.mdx: use async functions and proper ActionResult return type
- Add Warning about parameter name matching in basics.mdx
- Add Warning and Common Pitfalls section in add.mdx with clear examples
- Update AGENTS.md embedded docs to match
The agent injects special parameters by name matching, not by type.
Using incorrect parameter names (e.g., browser instead of browser_session)
causes tools to fail silently, which was a common user confusion.
Co-authored-by: mailmertunsal <mailmertunsal@gmail.com>
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Generate TOTP codes only when the placeholder key ends with
"bu_2fa_code". This supports multiple 2FA secrets and avoids false
matches in secret replacement.
<sup>Written for commit dd8a63ce41.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Updated aiohttp dependency to >=3.13.3 to fix conflicts with internal
libraries and ensure compatibility. Bumped package version to 0.11.3.
<sup>Written for commit e1ee8b753f.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Adds a menu-aware retry to agent reruns: if a menu item can’t be
matched, the agent re-opens the dropdown and retries quickly. This
reduces flaky failures in multi-level dropdown flows.
- **Bug Fixes**
- Detects menu openers (aria-haspopup, toggleSubMenu, expand-button,
role with aria-expanded).
- Detects menu items (menuitem roles, gw-action--inner, menu-related
classes, stable ax_name).
- Re-executes the opener step with a short delay and retries without
consuming a retry attempt.
- Improves logging and preserves existing backoff behavior.
- Adds unit tests for opener/item detection.
<sup>Written for commit 147db04c1c.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Added an Environment Variables section to the Agent parameters docs that
lists all timeout env vars with defaults and clear descriptions, plus
example usage. This lets users tune agent and browser behavior and
troubleshoot slow networks or complex pages without code changes.
<sup>Written for commit b71894b97b.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Made the Agent event bus stop timeout configurable via the
TIMEOUT_AgentEventBusStop env var (default 3s) to let tests and
deployments tune shutdown wait time. This connects to Linear 3076 by
allowing environments to adjust the idle wait instead of using a fixed
value.
<sup>Written for commit 9c042d7d98.
Summary will update on new commits.</sup>
<!-- End of auto-generated description by cubic. -->
Laith Weinberger
