JiuwenSwarm is an intelligent AI Agent built on openJiuwen. It extends the powerful capabilities of large language models directly to your fingertips through various communication apps you use daily.
Python
# Copyright (c) Huawei Technologies Co., Ltd. 2026. All rights reserved.
"""Unit tests for ``jiuwenswarm skill-train`` (jiuwenswarm.cli.skill_train)."""
from __future__ import annotations
import json
import sys
from types import SimpleNamespace
from unittest.mock import patch
import pytest
from jiuwenswarm.cli import skill_train as st
openjiuwen_skill_train = pytest.importorskip(
"openjiuwen.agent_evolving.skill_train",
reason="openjiuwen with agent_evolving.skill_train is required",
)
def _capture_logger_info(monkeypatch):
"""Record ``logger.info`` messages even when handlers bypass pytest caplog/capsys."""
recorded: list[str] = []
original = st.logger.info
def _info(msg, *args, **kwargs):
recorded.append(msg % args if args else str(msg))
return original(msg, *args, **kwargs)
monkeypatch.setattr(st.logger, "info", _info)
return recorded
class _StubTrainLaunchOptions:
"""Minimal stand-in so CLI unit tests do not require the launch API in openjiuwen."""
def __init__(self, **kwargs):
for key, value in kwargs.items():
setattr(self, key, value)
@property
def is_exec_backend(self) -> bool:
return getattr(self, "target_backend", "") != "openai_chat"
def resolved_backend(self) -> str:
return getattr(self, "target_backend", "")
@pytest.fixture
def stub_launch_api(monkeypatch):
"""Ensure TrainLaunchOptions / run_offline_* exist for CLI imports (CI may lack them)."""
monkeypatch.setattr(
openjiuwen_skill_train, "TrainLaunchOptions", _StubTrainLaunchOptions, raising=False,
)
# run_skill_train imports both names; install placeholders so ImportError cannot fire.
monkeypatch.setattr(
openjiuwen_skill_train, "run_offline_training", lambda opts: None, raising=False,
)
monkeypatch.setattr(
openjiuwen_skill_train, "run_offline_eval",
lambda opts, *, split: {}, raising=False,
)
return openjiuwen_skill_train
class TestParser:
@staticmethod
def test_defaults_target_jiuwenswarm_exec(monkeypatch):
monkeypatch.delenv("TARGET_BACKEND", raising=False)
monkeypatch.delenv("SKILL_TRAIN_ENV", raising=False)
args = st.build_parser().parse_args([])
assert args.env == "searchqa"
assert args.backend == "jiuwenswarm_cli_exec"
assert args.eval_only is False
assert args.no_trace_to_optimizer is False
@staticmethod
def test_options_map_to_launch_options(stub_launch_api, monkeypatch):
monkeypatch.delenv("JIUWENSWARM_CLI_PATH", raising=False)
args = st.build_parser().parse_args([
"--env", "docvqa", "--limit", "3", "--workers", "2", "--exec-timeout", "90",
"--gateway-url", "ws://gw:1/tui", "--name", "inst", "--mode", "code.plan",
"--cli-path", "bin/jiuwenswarm", "--no-gate", "--no-trace-to-optimizer",
"--data-root", "./data", "--optimizer-model", "opt",
])
opts = st._build_options(args)
assert opts.env_name == "docvqa"
assert opts.target_backend == "jiuwenswarm_cli_exec"
assert opts.limit == 3 and opts.workers == 2 and opts.exec_timeout == 90
assert opts.jiuwenswarm_gateway_url == "ws://gw:1/tui"
assert opts.jiuwenswarm_instance_name == "inst"
assert opts.jiuwenswarm_chat_mode == "code.plan"
assert opts.jiuwenswarm_cli_path == "bin/jiuwenswarm"
assert opts.use_gate is False
assert opts.jiuwenswarm_trace_to_optimizer is False
assert opts.data_root == "./data"
assert opts.optimizer_model == "opt"
assert opts.skill_inits is None
assert opts.is_exec_backend
@staticmethod
def test_skill_init_named_maps_to_skill_inits(stub_launch_api, monkeypatch):
monkeypatch.delenv("JIUWENSWARM_CLI_PATH", raising=False)
args = st.build_parser().parse_args([
"--skill-init", "legacy.md",
"--skill-init-named", "search=path/to/search/SKILL.md",
"--skill-init-named", "browse=path/to/browse/SKILL.md",
])
assert args.skill_init_named == [
("search", "path/to/search/SKILL.md"),
("browse", "path/to/browse/SKILL.md"),
]
opts = st._build_options(args)
assert opts.skill_init == ""
assert opts.skill_inits == {
"search": "path/to/search/SKILL.md",
"browse": "path/to/browse/SKILL.md",
}
@staticmethod
def test_parse_skill_init_named_rejects_bad_format():
with pytest.raises(SystemExit):
st.build_parser().parse_args(["--skill-init-named", "no-equals"])
@staticmethod
def test_cli_path_prefers_sibling_script(tmp_path, monkeypatch):
monkeypatch.delenv("JIUWENSWARM_CLI_PATH", raising=False)
fake_python = tmp_path / "Scripts" / "python.exe"
fake_python.parent.mkdir()
fake_python.write_text("")
(tmp_path / "Scripts" / "jiuwenswarm.exe").write_text("")
monkeypatch.setattr(sys, "executable", str(fake_python))
assert st._resolve_cli_path("") == str(tmp_path / "Scripts" / "jiuwenswarm.exe")
monkeypatch.setenv("JIUWENSWARM_CLI_PATH", "custom")
assert st._resolve_cli_path("") == "custom"
assert st._resolve_cli_path("explicit") == "explicit"
class TestGatewayCheck:
@staticmethod
def test_check_gateway_ok():
async def _mock_connect(url, **kwargs):
class FakeWs:
async def recv(self):
return json.dumps({"type": "event", "event": "connection.ack", "payload": {}})
async def close(self):
pass
return FakeWs()
with patch("jiuwenswarm.cli.gateway_client._connect_ws", _mock_connect):
assert st.check_gateway("ws://127.0.0.1:19001/tui") is True
@staticmethod
def test_check_gateway_refused():
async def _mock_connect(url, **kwargs):
raise ConnectionRefusedError("refused")
with patch("jiuwenswarm.cli.gateway_client._connect_ws", _mock_connect):
assert st.check_gateway("ws://127.0.0.1:1/tui", timeout=0.5) is False
class TestRun:
@staticmethod
def test_gateway_down_returns_3(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args(["--env", "searchqa"])
monkeypatch.setattr(st, "check_gateway", lambda url, **kw: False)
called = []
monkeypatch.setattr(
stub_launch_api, "run_offline_training", lambda opts: called.append(opts), raising=False,
)
assert st.run_skill_train(args) == st.EXIT_GATEWAY_UNREACHABLE
assert called == []
@staticmethod
def test_training_success(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args(["--env", "searchqa", "--skip-gateway-check", "--limit", "1"])
seen = {}
infos = _capture_logger_info(monkeypatch)
def _fake_train(opts):
seen["opts"] = opts
return SimpleNamespace(best_score=0.75, output_dir="out/x", best_skills={"default": "body"})
monkeypatch.setattr(stub_launch_api, "run_offline_training", _fake_train, raising=False)
assert st.run_skill_train(args) == 0
assert seen["opts"].limit == 1
assert seen["opts"].target_backend == "jiuwenswarm_cli_exec"
assert any("best_score=0.7500" in msg for msg in infos)
assert any("best_skill.md" in msg.replace("\\", "/") for msg in infos)
@staticmethod
def test_training_success_logs_multi_skill_paths(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args([
"--skip-gateway-check",
"--skill-init-named", "a=a.md",
"--skill-init-named", "b=b.md",
])
infos = _capture_logger_info(monkeypatch)
def _fake_train(opts):
assert opts.skill_inits == {"a": "a.md", "b": "b.md"}
return SimpleNamespace(
best_score=0.5,
output_dir="out/multi",
best_skills={"a": "A", "b": "B"},
)
monkeypatch.setattr(stub_launch_api, "run_offline_training", _fake_train, raising=False)
assert st.run_skill_train(args) == 0
joined = "\n".join(infos).replace("\\", "/")
assert "best_skills/a/SKILL.md" in joined
assert "best_skills/b/SKILL.md" in joined
@staticmethod
def test_eval_only(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args(["--eval-only", "--split", "val", "--skip-gateway-check"])
seen = {}
infos = _capture_logger_info(monkeypatch)
def _fake_eval(opts, *, split):
seen["split"] = split
return {
"env": "searchqa",
"target_backend": "jiuwenswarm_cli_exec",
"n_items": 2,
"hard": 0.5,
"soft": 0.6,
"output_dir": "o",
}
monkeypatch.setattr(stub_launch_api, "run_offline_eval", _fake_eval, raising=False)
assert st.run_skill_train(args) == 0
assert seen["split"] == "val"
assert any("Eval complete" in msg for msg in infos)
@staticmethod
def test_bad_config_returns_2(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args(["--skip-gateway-check"])
def _raise(opts):
raise FileNotFoundError("split_dir not found")
monkeypatch.setattr(stub_launch_api, "run_offline_training", _raise, raising=False)
assert st.run_skill_train(args) == st.EXIT_BAD_ARGS
@staticmethod
def test_chat_backend_skips_gateway_probe(stub_launch_api, monkeypatch):
args = st.build_parser().parse_args(["--backend", "openai_chat"])
monkeypatch.setattr(st, "check_gateway", lambda url, **kw: pytest.fail("must not probe"))
monkeypatch.setattr(
stub_launch_api,
"run_offline_training",
lambda opts: SimpleNamespace(best_score=1.0, output_dir="o"),
raising=False,
)
assert st.run_skill_train(args) == 0