Your lightweight open-source personal agent.
Python agent-frameworkai-agentai-agentschatbotchatopsdiscord-botllm-agentsllmslocal-llmmcpmodel-context-protocolmulti-agentopenai-compatibleopenclawpersonal-ai-assistantpythonself-hostedtelegram-bot-ai-assistantwebuiworkflow-automation
fix(web): propagate web_fetch failures as structured tool errors
Kail Tian4f7479ca64ea
diff --git a/nanobot/agent/tools/web.py b/nanobot/agent/tools/web.pyindex 087acf04c..0fc75297b 100644--- a/nanobot/agent/tools/web.py+++ b/nanobot/agent/tools/web.py@@ -1166,7 +1166,7 @@ class WebFetchTool(Tool):max_chars = cast(int, kwargs.pop("maxChars", max_chars) or self.max_chars)is_valid, error_msg = _validate_url_safe(url)if not is_valid:- return json.dumps({"error": f"URL validation failed: {error_msg}", "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": f"URL validation failed: {error_msg}", "url": url}, ensure_ascii=False))# Detect and fetch images directly to avoid Jina's textual image captioning.# This local preflight also proves that no credential-bearing URL occurs@@ -1184,9 +1184,9 @@ class WebFetchTool(Tool):))if redirect_error:- return json.dumps({"error": redirect_error, "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": redirect_error, "url": url}, ensure_ascii=False))if r is None:- return json.dumps({"error": "Fetch failed", "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": "Fetch failed", "url": url}, ensure_ascii=False))jina_remote_safe = not chain_carries_credentialstry:@@ -1201,7 +1201,7 @@ class WebFetchTool(Tool):except Exception as e:unsafe_error = _unsafe_url_request_error(e)if unsafe_error is not None:- return json.dumps({"error": f"URL validation failed: {unsafe_error}", "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": f"URL validation failed: {unsafe_error}", "url": url}, ensure_ascii=False))logger.debug("Pre-fetch image detection failed for {} ({})",_redact_url_for_log(url),@@ -1277,9 +1277,9 @@ class WebFetchTool(Tool):headers={"User-Agent": self.user_agent},)if redirect_error:- return json.dumps({"error": redirect_error, "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": redirect_error, "url": url}, ensure_ascii=False))if r is None:- return json.dumps({"error": "Fetch failed", "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": "Fetch failed", "url": url}, ensure_ascii=False))r.raise_for_status()ctype = r.headers.get("content-type", "")@@ -1318,14 +1318,14 @@ class WebFetchTool(Tool):_redact_url_for_log(url),type(e).__name__,)- return json.dumps({"error": f"Proxy error: {e}", "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": f"Proxy error: {e}", "url": url}, ensure_ascii=False))except Exception as e:logger.warning("WebFetch error for {} ({})",_redact_url_for_log(url),type(e).__name__,)- return json.dumps({"error": str(e), "url": url}, ensure_ascii=False)+ return ToolResult.error(json.dumps({"error": str(e), "url": url}, ensure_ascii=False))def _extract_readable_html(self, html_content: str, extract_mode: str) -> str:from readability import Document # pyright: ignore[reportMissingTypeStubs]diff --git a/tests/agent/test_runner_errors.py b/tests/agent/test_runner_errors.pyindex 3211014f7..de2c8b981 100644--- a/tests/agent/test_runner_errors.py+++ b/tests/agent/test_runner_errors.py@@ -3,6 +3,8 @@ session message isolation, and tool result preservation."""from __future__ import annotations+import json+import socketfrom types import SimpleNamespacefrom unittest.mock import AsyncMock, MagicMock@@ -12,6 +14,8 @@ from agent.runner_helpers import make_run_specfrom nanobot.agent.hook import AgentHook, AgentHookContextfrom nanobot.agent.tools import ToolResultfrom nanobot.agent.tools.execution import execute_tool_calls+from nanobot.agent.tools.registry import ToolRegistry+from nanobot.agent.tools.web import WebFetchToolfrom nanobot.config.schema import AgentDefaultsfrom nanobot.providers.base import LLMProvider, LLMResponse, ToolCallRequest@@ -237,6 +241,76 @@ async def test_runner_returns_structured_tool_error_to_model_for_recovery():][email protected]("url,ssrf", [+ pytest.param("not-a-url", False, id="invalid-url"),+ pytest.param("http://127.0.0.1/admin", True, id="blocked-loopback"),+])+async def test_runner_recognizes_web_fetch_errors(monkeypatch, url, ssrf):+ from nanobot.agent.runner import AgentRunner++ monkeypatch.setattr("nanobot.security.network._allowed_networks", [])+ monkeypatch.setattr("nanobot.security.network.socket.getaddrinfo", MagicMock(+ return_value=[(socket.AF_INET, socket.SOCK_STREAM, 0, "", ("127.0.0.1", 0))],+ ))+ http_client = MagicMock(side_effect=AssertionError("Rejected URLs must not reach HTTP"))+ monkeypatch.setattr("nanobot.agent.tools.web.httpx.AsyncClient", http_client)+ tools = ToolRegistry()+ tools.register(WebFetchTool())+ hook = AgentHook()+ error_hook = AsyncMock()+ success_hook = AsyncMock()+ monkeypatch.setattr(hook, "on_execute_tool_error", error_hook)+ monkeypatch.setattr(hook, "after_execute_tool", success_hook)+ provider = MagicMock(spec=LLMProvider)+ provider.chat_stream_with_retry = AsyncMock(side_effect=[+ LLMResponse(content="fetching", tool_calls=[+ ToolCallRequest(id="fetch_1", name="web_fetch", arguments={"url": url}),+ ]),+ LLMResponse(content="Unable to fetch this URL."),+ ])++ result = await AgentRunner().run(make_run_spec(provider,+ initial_messages=[{"role": "user", "content": f"Read {url}"}],+ tools=tools,+ model="test-model",+ max_iterations=2,+ max_tool_result_chars=_MAX_TOOL_RESULT_CHARS,+ hook=hook,+ ))++ assert result.stop_reason == "completed"+ assert result.final_content == "Unable to fetch this URL."+ assert len(result.tool_events) == 1+ assert result.tool_events[0]["status"] == "error"+ assert result.tools_used == []+ error_hook.assert_awaited_once()+ success_hook.assert_not_awaited()+ http_client.assert_not_called()+ raw_error = error_hook.await_args.args[-1]+ assert isinstance(raw_error, ToolResult) and raw_error.is_error+ raw_payload = json.loads(raw_error)+ assert raw_payload["url"] == url+ assert raw_payload["error"].startswith("URL validation failed:")++ assert provider.chat_stream_with_retry.await_count == 2+ next_request = provider.chat_stream_with_retry.await_args_list[1].kwargs["messages"]+ tool_message = next(message for message in next_request if message.get("role") == "tool")+ assert tool_message["tool_call_id"] == "fetch_1"+ content = tool_message["content"]+ assert content.startswith(str(raw_error) + "\n\n")+ assert json.JSONDecoder().raw_decode(content)[0] == raw_payload+ retry_hint = "[Analyze the error above and try a different approach.]"+ if ssrf:+ assert result.tool_events[0]["detail"].startswith("ssrf_violation:")+ assert "non-bypassable security boundary" in content+ assert "Do not retry" in content+ assert "tools.ssrfWhitelist" in content+ assert retry_hint not in content+ else:+ assert content.count(retry_hint) == 1++@pytest.mark.asyncioasync def test_runner_preserves_successful_exec_output_that_starts_with_error():from nanobot.agent.runner import AgentRunnerdiff --git a/tests/tools/test_web_fetch_security.py b/tests/tools/test_web_fetch_security.pyindex 9d57e70a7..63097d4d0 100644--- a/tests/tools/test_web_fetch_security.py+++ b/tests/tools/test_web_fetch_security.py@@ -5,13 +5,14 @@ from __future__ import annotationsimport asyncioimport jsonimport socket-from unittest.mock import patch+from unittest.mock import AsyncMock, patchfrom urllib.request import getproxies_environmentimport httpximport pytestfrom nanobot.agent.tools import web as web_module+from nanobot.agent.tools.registry import is_tool_error_resultfrom nanobot.agent.tools.web import WebFetchTool, _get_with_safe_redirectsfrom nanobot.config.schema import WebFetchConfigfrom nanobot.security.network import PinnedDNSAsyncTransport@@ -96,6 +97,7 @@ async def test_web_fetch_blocks_private_ip():with patch("nanobot.security.network.socket.getaddrinfo", _fake_resolve_private):result = await tool.execute(url="http://169.254.169.254/computeMetadata/v1/")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in dataassert "private" in data["error"].lower() or "blocked" in data["error"].lower()@@ -108,6 +110,7 @@ async def test_web_fetch_blocks_localhost():with patch("nanobot.security.network.socket.getaddrinfo", _resolve_localhost):result = await tool.execute(url="http://localhost/admin")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in data@@ -126,6 +129,7 @@ async def test_web_fetch_blocks_localhost_even_in_full_workspace_scope(tmp_path)finally:reset_workspace_scope(token)data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in data@@ -139,6 +143,7 @@ async def test_web_fetch_result_contains_untrusted_flag(monkeypatch: pytest.Monkresult = await tool.execute(url="https://example.com/page")data = json.loads(result)+ assert not is_tool_error_result(result)assert data.get("untrusted") is Trueassert "[External content" in data.get("text", "")@@ -278,6 +283,7 @@ async def test_web_fetch_does_not_fallback_after_pinned_dns_rebind_rejection(monresult = await tool.execute(url="http://evil.example/page")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in dataassert "blocked" in data["error"].lower()assert calls["evil.example"] == 3@@ -452,6 +458,7 @@ async def test_web_fetch_blocks_private_redirect_before_readability_request(monkresult = await tool.execute(url="https://attacker.example/start")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in dataassert "redirect blocked" in data["error"].lower()assert requested == ["https://attacker.example/start"]@@ -498,6 +505,7 @@ async def test_web_fetch_blocks_private_redirect_before_returning_image(monkeyparesult = await tool.execute(url="https://example.com/image.png")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in dataassert "redirect blocked" in data["error"].lower()@@ -539,6 +547,55 @@ async def test_web_fetch_does_not_request_private_redirect_target(monkeypatch):result = await tool.execute(url="https://attacker.example/start")data = json.loads(result)+ assert is_tool_error_result(result)assert "error" in dataassert "redirect blocked" in data["error"].lower()assert requested == ["https://attacker.example/start"]+++ ("failure", "error_text"),+ [("http", "404"), ("timeout", "request timed out"), ("proxy", "Proxy error: proxy unavailable")],+)+async def test_web_fetch_marks_terminal_fetch_failures_as_errors(monkeypatch, failure, error_text):+ url = "https://93.184.216.34/page"+ tool = WebFetchTool(config=WebFetchConfig(use_jina_reader=False))++ def handler(request: httpx.Request) -> httpx.Response:+ if failure == "timeout":+ raise httpx.ReadTimeout("request timed out", request=request)+ if failure == "proxy":+ raise httpx.ProxyError("proxy unavailable", request=request)+ return httpx.Response(404, request=request)++ monkeypatch.setattr(web_module, "_pinned_dns_transport", lambda: httpx.MockTransport(handler))++ result = await tool.execute(url=url)++ assert is_tool_error_result(result)+ data = json.loads(result)+ assert data["url"] == url+ assert error_text in data["error"]+++async def test_web_fetch_jina_fallback_keeps_json_error_content_successful(monkeypatch):+ url = "https://93.184.216.34/page"+ tool = WebFetchTool()+ jina = AsyncMock(return_value=None)+ source = {"error": "This is fetched document content, not a tool failure."}++ def handler(request: httpx.Request) -> httpx.Response:+ return httpx.Response(200, json=source, request=request)++ monkeypatch.setattr(web_module, "_pinned_dns_transport", lambda: httpx.MockTransport(handler))+ monkeypatch.setattr(tool, "_fetch_jina", jina)++ result = await tool.execute(url=url)++ jina.assert_awaited_once()+ assert not is_tool_error_result(result)+ data = json.loads(result)+ assert data["status"] == 200+ assert data["extractor"] == "json"+ assert data["untrusted"] is True+ assert json.dumps(source, indent=2) in data["text"]