Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

Record single-site collection proof

Magnus Müllerecb41c3094c4
diff --git a/RUST_AGENT_INTEGRATION_FEATURES.md b/RUST_AGENT_INTEGRATION_FEATURES.md
index 5dd48d677..65af42855 100644
--- a/RUST_AGENT_INTEGRATION_FEATURES.md
+++ b/RUST_AGENT_INTEGRATION_FEATURES.md
@@ -1227,6 +1227,12 @@ Terminal core branch: `magnus/browser-use-rust-main-integration` at terminal mai
- This targets the 2026-06-04 five-task gate failure where the AUD networking-pricing task spent many turns varying one missing access-point search instead of moving through the remaining requested procurement table.
- Proof: terminal `system_prompt_bounds_multi_item_collection_loops`, terminal `cargo fmt --check -p browser-use-agent`, and terminal `cargo build -q -p browser-use-cli`.
+242. Rust Agent single-site collection parity
+ - The terminal browser-agent prompt now gives general single-site guidance for tasks that request one website, vendor, domain, or source.
+ - The agent is instructed to pick one viable domain early, finish the collection checklist on that domain, avoid stitching rows across domains, and mark missing items unavailable for the committed domain instead of vendor-hopping.
+ - This targets the 2026-06-04 focused task-4 diagnostic where the agent spent 21 steps scouting vendor domains without committing to one AUD source.
+ - Proof: terminal `system_prompt_commits_single_site_collection_to_one_domain`, terminal `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent system_prompt_ -- --nocapture`, terminal `cargo fmt --check -p browser-use-agent`, and terminal `cargo build -q -p browser-use-cli`.
+
## Current Verification
- `python3 -m py_compile browser_use/agent/service.py browser_use/rust/service.py browser_use/rust/__init__.py browser_use/__init__.py browser_use/llm/models.py tests/ci/test_rust_agent.py tests/ci/models/test_llm_model_factory.py examples/rust_agent/basic.py examples/rust_agent/real_v8_smoke.py`
@@ -1250,6 +1256,7 @@ Terminal core branch: `magnus/browser-use-rust-main-integration` at terminal mai
- `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent selected_remote_cdp_rewrites_wrong_browser_family_commands -- --nocapture`
- `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent browser_mode_instruction_guides_remote_cdp_to_direct_page_work -- --nocapture`
- `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent system_prompt_bounds_multi_item_collection_loops -- --nocapture`
+- `CARGO_INCREMENTAL=0 cargo test -q -p browser-use-agent system_prompt_ -- --nocapture`
- `cargo fmt --check -p browser-use-agent -p browser-use-browser` on terminal branch `magnus/browser-use-rust-main-integration`
- `cargo fmt --check -p browser-use-agent` on terminal branch `magnus/browser-use-rust-main-integration`
- `python3 -m py_compile crates/browser-use-browser/src/browser_script_helpers.py` on terminal branch `magnus/browser-use-rust-main-integration`
@@ -1419,6 +1426,9 @@ Terminal core branch: `magnus/browser-use-rust-main-integration` at terminal mai
- Scores were task 1 `0.9`, task 2 `0.5`, task 3 `0.9`, task 4 `0.0`, task 5 `1.0`; mean score `0.66`, total batch time `891.75s`.
- Validated that Remote CDP setup no longer wastes early rejected connect steps and continuous `.capture.frames` directories are not created by default.
- Did not justify scaling to 50: task 4 reached 50 steps and 867.84s after repeatedly varying one missing product search; the terminal child was stopped to avoid burning the full timeout.
+- 2026-06-04 focused real_v8 task-4 Agent SDK diagnostic:
+ - Run `kh760nrdkqwsjxb539c6v740m58808s1` used `BU_RUNTIME=brust`, `--browser browser-use-cloud`, `--judge-type agent_sdk`, `--model claude-sonnet-4-6`, `--eval-model claude-sonnet-4-6`, `--parallel-runs 1`, and `--max-steps 75`.
+ - Score was `0.0` after the terminal child was stopped at 271.88s and 21 steps; the run showed vendor-domain scouting without committing to one AUD source, which feature 242 addresses.
- 2026-06-03 three-task real_v8 cloud smoke through the Python `Agent` API and Rust terminal core launched and returned for real_v8 `18`, `14`, and `20` with `DEFAULT_LLM=openai_gpt_4_1_mini`, `BROWSER_USE_RUST_BROWSER_MODE=cloud`, and the local debug `browser-use-terminal`, but it is not a clean benchmark pass:
- real_v8 `18`: returned `successful=true`, but final output was `[empty]` after failing to verify AND Digital leadership information and included generated browser-script JavaScript quoting errors.
- real_v8 `14`: returned `successful=true`, but final output said ycombinator.com was blocked/unresponsive and did not extract the requested Winter 2025 B2B company data.