langgenius · crazywoola · Sep 15, 2026 · Sep 4, 2026 · Sep 8, 2026 · Sep 9, 2026
diff --git a/tools/duckduckgo/README.md b/tools/duckduckgo/README.md
@@ -26,4 +26,17 @@ Both Chatflow and Workflow applications support adding DuckDuckGo tool nodes. Tw
 
 #### - Agent applications
 
-Add the DuckDuckGo tool in the Agent application, then enter the search command to call this tool.
+Add the DuckDuckGo tool in the Agent application, then enter the search command to call this tool.
+
+## Rate limiting and blocked hosts
+
+`ddgs` scrapes public search engines rather than calling an API, so a host that runs many searches in a short window (for example a workflow that loops over several search nodes) gets rate-limited or served captcha pages. When that happens the tool raises `DuckDuckGo text search returned nothing for '...' after 3 attempts`.
+
+The plugin already spaces searches 2 seconds apart per process and retries with exponential backoff. If you still hit the error:
+
+- **Reduce volume.** Merge overlapping queries into one search node and lower loop counts.
+- **Pin `Search engines`.** Set it to a comma-separated list of engines that still answer from your host, e.g. `duckduckgo,brave,yahoo`. Leave empty to rotate through all of them.
+- **Set `Proxy server`.** Route through an http/https/socks proxy (rotating proxies work best), or `tb` for a local Tor Browser.
+- **Handle failures in the workflow.** Enable *Retry on failure* and an *Error handling* strategy on the tool node so one blocked search degrades instead of aborting the run.
+
+For sustained high-volume use, an API-backed search plugin (Tavily, Serper, Brave Search, SearXNG) is more reliable than scraping.
diff --git a/tools/duckduckgo/manifest.yaml b/tools/duckduckgo/manifest.yaml
@@ -1,4 +1,4 @@
-version: 0.0.12
+version: 0.0.13
 type: plugin
 author: "langgenius"
 name: "duckduckgo"

diff --git a/tools/duckduckgo/tests/__init__.py b/tools/duckduckgo/tests/__init__.py
diff --git a/tools/duckduckgo/tests/test_ddgs_hardening.py b/tools/duckduckgo/tests/test_ddgs_hardening.py
@@ -0,0 +1,145 @@
+"""Tests for tools/ddgs_hardening.py. Need the real ``ddgs`` package (they patch its classes);
+skipped when only the stub from test_ddgs_utils is present."""
+
+from __future__ import annotations
+
+import sys
+import types
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+ddgs = pytest.importorskip("ddgs")
+pytest.importorskip("lxml")
+if isinstance(ddgs, types.ModuleType) and getattr(ddgs, "__file__", None) is None:
+    # the in-memory stub installed by test_ddgs_utils when ddgs is not installed
+    pytest.skip("real ddgs package required", allow_module_level=True)
+pytest.importorskip("ddgs.http_client")
+
+from ddgs.base import BaseSearchEngine  # noqa: E402
+from ddgs.engines.yahoo import Yahoo  # noqa: E402
+from ddgs.exceptions import DDGSException  # noqa: E402
+
+from tools import ddgs_hardening  # noqa: E402
+
+ddgs_hardening.apply()
+
+
+class _Resp:
+    def __init__(self, status: int, text: str) -> None:
+        self.status_code, self.text = status, text
+
+
+class _FlakyClient:
+    """Drops the connection ``drops`` times, then answers 200."""
+
+    def __init__(self, drops: int, body: str = "<html>ok</html>") -> None:
+        self.drops, self.body, self.calls = drops, body, 0
+
+    def request(self, *a, **k):
+        self.calls += 1
+        if self.calls <= self.drops:
+            raise DDGSException(
+                "RequestError: RequestError('error sending request for url (https://search.yahoo.com/...) "
+                "> peer closed connection without sending TLS close_notify')"
+            )
+        return _Resp(200, self.body)
+
+
+@pytest.fixture
+def fast_sleep(monkeypatch):
+    monkeypatch.setattr(ddgs_hardening.time, "sleep", lambda s: None)
+
+
+def test_transport_drop_is_resent_within_the_same_engine_call(fast_sleep):
+    engine = Yahoo()
+    engine.http_client = _FlakyClient(drops=1)
+    assert BaseSearchEngine.request(engine, "GET", "https://search.yahoo.com/search") == "<html>ok</html>"
+    assert engine.http_client.calls == 2
+
+
+def test_transport_drop_gives_up_after_configured_retries(fast_sleep, monkeypatch):
+    monkeypatch.setattr(ddgs_hardening, "TRANSPORT_RETRIES", 2)
+    engine = Yahoo()
+    engine.http_client = _FlakyClient(drops=5)
+    with pytest.raises(DDGSException):
+        BaseSearchEngine.request(engine, "GET", "https://search.yahoo.com/search")
+    assert engine.http_client.calls == 3  # 1 + 2 retries
+
+
+def test_non_transport_errors_are_not_retried(fast_sleep):
+    class _Client:
+        calls = 0
+
+        def request(self, *a, **k):
+            self.calls += 1
+            raise DDGSException("proxy refused the connection: 407")
+
+    engine = Yahoo()
+    engine.http_client = _Client()
+    with pytest.raises(DDGSException):
+        BaseSearchEngine.request(engine, "GET", "https://search.yahoo.com/search")
+    assert engine.http_client.calls == 1
+
+
+@pytest.mark.parametrize(
+    ("engine", "status", "body", "expected"),
+    [
+        ("brave", 429, "<html>captcha</html>", "429-ratelimit"),
+        ("mojeek", 403, "", "403-forbidden"),
+        ("duckduckgo", 202, "<html></html>", "202-ratelimit"),
+        ("yahoo", 200, "<html>We had temporary problems searching for web pages.</html>", "throttle-page"),
+        ("startpage", 200, "<html>please solve this captcha</html>", "captcha-page"),
+        ("yahoo", 200, "<div class='dd algo relsrch'>…</div>", "ok-desktop"),
+        ("yahoo", 200, "<section class='dd algo-sr'>…</section>", "ok-mobile-layout"),
+    ],
+)
+def test_classify(engine, status, body, expected):
+    assert ddgs_hardening.classify(engine, status, body) == expected
+
+
+MOBILE_PAGE = """
+<html><body>
+<section class="dd fst algo s-algo algo-sr Sr">
+  <div class="grp grp-talgo ag-1 u-tapHgt cur-p"><div class="compTitle p-r">
+    <h3 class="title d-ib mt-42"><a class="s-title fz-m" href="https://r.search.yahoo.com/_ylt=x/RU=https%3a%2f%2fen.wikipedia.org%2fwiki%2fNan_Fung_Group/RK=2/RS=y">
+      <span class="title-url">Wikipedia https://en.m.wikipedia.org</span>Nan Fung Group - Wikipedia</a></h3>
+  </div></div>
+  <div class="grp grp-talgo-ext"><p class="s-desc lh-20">Nan Fung Group is a privately held group of companies.</p></div>
+</section>
+<section class="dd lst algo s-algo algo-sr Sr">
+  <div class="grp grp-talgo ag-2"><div class="compTitle p-r">
+    <h3 class="title"><a class="s-title fz-m" href="https://r.search.yahoo.com/_ylt=y/RU=https%3a%2f%2fwww.nanfung.com/RK=2/RS=z">
+      <span class="title-url">Nan Fung Group https://www.nanfung.com</span>Home - Nan Fung Group</a></h3>
+  </div></div>
+  <p class="s-desc">Founded in 1954.</p>
+</section>
+</body></html>
+"""
+
+
+def test_yahoo_mobile_layout_is_parsed():
+    results = Yahoo().extract_results(MOBILE_PAGE)
+    assert [r.title for r in results] == ["Nan Fung Group - Wikipedia", "Home - Nan Fung Group"]
+    assert results[0].body.startswith("Nan Fung Group is a privately held")
+    assert "/RU=" in results[0].href
+
+
+def test_desktop_layout_still_uses_the_stock_parser():
+    page = "<html><div class='dd algo relsrch'><div class='compTitle Title'><h3>T</h3><a href='https://r.search.yahoo.com/RU=https%3a%2f%2fx.com/RK=2/RS=1'>t</a></div><div class='compText Text'>B</div></div></html>"
+    results = Yahoo().extract_results(page)
+    assert len(results) == 1 and results[0].title == "T"
+
+
+def test_cooldown_excludes_blocked_engines(monkeypatch):
+    monkeypatch.setattr(ddgs_hardening, "_cooldown_until", {})
+    monkeypatch.setattr(ddgs_hardening, "ENGINE_COOLDOWN_SECONDS", 60.0)
+    assert ddgs_hardening.healthy_backend("text") is None  # nothing blocked -> leave 'auto'
+    ddgs_hardening._record("q", "brave", 429, 10, "429-ratelimit")
+    ddgs_hardening._record("q", "google", 429, 10, "429-ratelimit")
+    healthy = ddgs_hardening.healthy_backend("text")
+    assert healthy is not None
+    assert "brave" not in healthy.split(",") and "google" not in healthy.split(",")
+    assert "yahoo" in healthy.split(",")