fix(skills): decode skill-review analyzer output as UTF-8 (#7432)

The analyzer prints ensure_ascii=False JSON, but the parent pipes used
the ambient locale codec and the child inherited the ambient stdio
codec: non-ASCII skill facts crashed the review pass on cp1252
Windows and mojibake'd on cp936. Pin both sides to UTF-8.

Fixes #7387

backend/tests/test_review_changed_public_skills_utf8.py

@@ -0,0 +1,101 @@
+from __future__ import annotations
+
+import importlib.util
+import os
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+PAYLOAD = {"summary": {"blockers": 0, "errors": 0, "warnings": 0, "infos": 1}, "completeness": {}, "notes": ["判定说明 ✗ 🎉"]}
+
+DRIVER_SOURCE = """
+import importlib
+import sys
+from pathlib import Path
+
+scripts_dir, repo_root = sys.argv[1], Path(sys.argv[2])
+sys.path.insert(0, scripts_dir)
+module = importlib.import_module("review_changed_public_skills")
+facts = module.collect_review_facts(repo_root / "skills" / "public" / "demo", repo_root, sys.executable)
+expected = {notes!r}
+if facts is None or facts.get("notes") != expected:
+    print("MISMATCH:", facts)
+    sys.exit(1)
+print("OK")
+""".format(notes=PAYLOAD["notes"])
+
+
+@pytest.fixture
+def analyzer_stub(tmp_path: Path) -> Path:
+    """A repo-root stand-in whose `deerflow.skills.review.cli` prints raw UTF-8 JSON."""
+    stub_root = tmp_path / "repo_root"
+    package = stub_root / "deerflow" / "skills" / "review"
+    package.mkdir(parents=True)
+    for parent in (stub_root / "deerflow", stub_root / "deerflow" / "skills", package):
+        (parent / "__init__.py").write_text("", encoding="utf-8")
+    (package / "cli.py").write_text(
+        "import json\nprint(json.dumps(" + repr(PAYLOAD) + ", ensure_ascii=False))\n",
+        encoding="utf-8",
+    )
+    return stub_root
+
+
+def _load_script_module():
+    scripts_dir = REPO_ROOT / "scripts"
+    spec = importlib.util.spec_from_file_location("review_changed_public_skills_under_test", scripts_dir / "review_changed_public_skills.py")
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    sys.modules[spec.name] = module
+    sys.path.insert(0, str(scripts_dir))
+    try:
+        spec.loader.exec_module(module)
+    finally:
+        sys.path.remove(str(scripts_dir))
+    return module
+
+
+def test_collect_review_facts_round_trips_non_ascii_analyzer_output(analyzer_stub: Path, tmp_path: Path):
+    """The analyzer emits ensure_ascii=False JSON, so both pipe directions must be UTF-8.
+
+    The driver runs with UTF-8 mode disabled and a C locale so the ambient codec is
+    a non-UTF-8 one (ASCII on Linux, the ANSI code page on Windows); the assertion
+    fails before the UTF-8 pinning and passes after it, on every platform.
+    """
+    driver = tmp_path / "driver.py"
+    driver.write_text(DRIVER_SOURCE, encoding="utf-8")
+
+    env = {
+        "PATH": os.environ.get("PATH", ""),
+        "SYSTEMROOT": os.environ.get("SYSTEMROOT", ""),
+        "PYTHONUTF8": "0",
+        "LC_ALL": "C",
+        "PYTHONCOERCECLOCALE": "0",
+        "PYTHONDONTWRITEBYTECODE": "1",
+    }
+    result = subprocess.run(
+        [sys.executable, str(driver), str(REPO_ROOT / "scripts"), str(analyzer_stub)],
+        env=env,
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        errors="replace",
+        check=False,
+        timeout=120,
+    )
+
+    assert result.returncode == 0, result.stdout + result.stderr
+    assert "OK" in result.stdout
+
+
+def test_review_env_pins_analyzer_stdio_to_utf8(analyzer_stub: Path, monkeypatch):
+    """review_env must override any inherited PYTHONUTF8/PYTHONIOENCODING for the analyzer."""
+    module = _load_script_module()
+    monkeypatch.setenv("PYTHONIOENCODING", "cp1252")
+    monkeypatch.setenv("PYTHONUTF8", "0")
+    env = module.review_env(analyzer_stub)
+    assert env["PYTHONIOENCODING"] == "utf-8"
+    assert env["PYTHONUTF8"] == "1"

scripts/review_changed_public_skills.py

@@ -318,6 +318,7 @@ def collect_review_facts(package: Path, repo_root: Path, python_executable: str)
         env=review_env(repo_root),
         capture_output=True,
         text=True,
+        encoding="utf-8",
         check=False,
     )
     if result.returncode != 0:
@@ -381,6 +382,11 @@ def review_env(repo_root: Path) -> dict[str, str]:
     harness_path = repo_root / "backend" / "packages" / "harness"
     existing_pythonpath = env.get("PYTHONPATH")
     env["PYTHONPATH"] = str(harness_path) if not existing_pythonpath else f"{harness_path}{os.pathsep}{existing_pythonpath}"
+    # The analyzer prints ensure_ascii=False JSON, so its stdio must be UTF-8
+    # regardless of the ambient locale, or non-ASCII skill facts crash the
+    # analyzer before the UTF-8-decoding parent ever sees them.
+    env["PYTHONIOENCODING"] = "utf-8"
+    env["PYTHONUTF8"] = "1"
     return env