完善 MCP/会话上下文治理并补齐搜索与诊断能力。

补充历史压缩与会话诊断链路，强化 tool pairing 与空响应兜底观测；完善 Admin MCP 导入/绑定与相关前端展示；新增 web_search_fast/web_fetch_clean 及多项测试与文档更新，并同步技能安装与角色绑定策略改进。

Made-with: Cursor

tests/test_web_fetch_clean_tool.py

@@ -0,0 +1,43 @@
+from __future__ import annotations
+
+from oclaw.runtime.tools.public.web_fetch_clean_tool import web_fetch_clean_tool
+
+
+class _Resp:
+    def __init__(self, body: bytes, *, status: int = 200, headers: dict[str, str] | None = None):
+        self._body = body
+        self.status = status
+        self.headers = headers or {"content-type": "text/html; charset=utf-8"}
+
+    def read(self) -> bytes:
+        return self._body
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return False
+
+
+def test_web_fetch_clean_html_strip(monkeypatch) -> None:
+    html = b"""
+    <html><head><title> Hello </title><style>.x{}</style></head>
+    <body><h1>Title</h1><p>Para <b>one</b>.</p><script>alert(1)</script></body></html>
+    """
+
+    monkeypatch.setattr("urllib.request.urlopen", lambda req, timeout=0: _Resp(html))  # noqa: ARG005
+    out = web_fetch_clean_tool().handler({"url": "https://example.com"})
+    assert out.get("ok") is True, out
+    assert out.get("title") == "Hello"
+    text = str(out.get("text") or "")
+    assert "Para" in text
+    assert "one" in text
+    assert "alert" not in str(out.get("text") or "")
+
+
+def test_web_fetch_clean_http_error(monkeypatch) -> None:
+    monkeypatch.setattr("urllib.request.urlopen", lambda req, timeout=0: _Resp(b"no", status=403))  # noqa: ARG005
+    out = web_fetch_clean_tool().handler({"url": "https://example.com"})
+    # Our stub doesn't raise HTTPError; status is checked and returns http_error.
+    assert out.get("ok") is False
+    assert out.get("error_code") == "http_error"