fix web fetch atom api recovery

tests/test_web_fetch_batch.py

@@ -1,7 +1,7 @@
 import asyncio
 import json
 
-from src.agent_tools.web_tools import WebFetchTool
+from src.agent_tools.web_tools import WebFetchTool, _arxiv_listing_api_hint
 from src.search import content as content_mod
 from src.tool_schemas import FUNCTION_TOOL_SCHEMAS, function_call_to_tool_block
 from src.clean_agent_preview import normalize_preview_function_args
@@ -79,6 +79,20 @@ def test_web_fetch_does_not_decode_local_html_or_binary_media(tmp_path):
     assert "inspect_media" in video["error"]
 
 
+def test_arxiv_list_406_offers_exact_public_api_recovery():
+    hint = _arxiv_listing_api_hint(
+        "https://arxiv.org/list/cs.CV?date=2026-02-25",
+        "HTTP 406: blocked",
+    )
+
+    assert "export.arxiv.org/api/query" in hint
+    assert "cat%3Acs.CV" in hint
+    assert "submittedDate%3A%5B202602250000" in hint
+    assert not _arxiv_listing_api_hint(
+        "https://example.com/list/cs.CV?date=2026-02-25", "HTTP 406"
+    )
+
+
 def test_web_fetch_schema_and_native_parser_accept_urls_batch():
     schema = next(
         item["function"] for item in FUNCTION_TOOL_SCHEMAS

tests/test_web_fetch_size_caps.py

@@ -115,6 +115,16 @@ def test_body_under_cap_is_untouched(monkeypatch, no_cache):
     assert r["fetched_bytes"] == len(b"hello world")
 
 
+def test_atom_xml_api_response_is_preserved_as_readable_evidence(monkeypatch, no_cache):
+    body = b"<?xml version='1.0'?><feed><entry><title>Daily paper</title></entry></feed>"
+    _patch_stream(monkeypatch, _FakeStream(body, content_type="application/atom+xml"))
+
+    result = content_mod.fetch_webpage_content("https://example.com/api/feed")
+
+    assert result["success"] is True
+    assert "<title>Daily paper</title>" in result["content"]
+
+
 def test_body_over_soft_cap_truncates_with_flags(monkeypatch, no_cache):
     body = b"x" * (WEB_FETCH_SOFT_MAX_BYTES + 50_000)
     _patch_stream(monkeypatch, _FakeStream(body, content_length=len(body)))