feat(tasks): add skill-templated task graph execution

app-instance/backend/tests/unit/test_agent_loop.py

@@ -1,8 +1,10 @@
 import asyncio
+import json
 from contextlib import suppress
 from typing import Any
 
 from beaver.engine import AgentLoop, AgentRunResult, EngineLoader
+from beaver.engine import loop as loop_module
 
 
 def _run_result(run_id: str, output_text: str) -> AgentRunResult:
@@ -45,3 +47,37 @@ def test_running_loop_handles_reentrant_submit_direct(tmp_path) -> None:
         assert calls == ["outer", "inner"]
 
     asyncio.run(run_case())
+
+
+def test_web_search_loop_guard_stops_after_repeated_low_quality_results() -> None:
+    guard = loop_module._WebSearchLoopGuard()
+    low_quality = json.dumps(
+        {
+            "success": True,
+            "query": "weather beijing",
+            "quality": "low",
+            "results": [{"title": "Example", "url": "https://example.com", "snippet": ""}],
+        }
+    )
+
+    assert guard.observe_result("web_search", low_quality) is None
+    assert guard.observe_result("web_search", low_quality) is None
+
+    guidance = guard.observe_result("web_search", low_quality)
+
+    assert guidance is not None
+    assert guidance["finish_reason"] == "web_search_low_quality_budget"
+    assert "weather beijing" in guidance["message"]
+
+
+def test_web_search_loop_guard_resets_after_useful_result() -> None:
+    guard = loop_module._WebSearchLoopGuard()
+    low_quality = json.dumps({"success": True, "query": "weather", "quality": "low", "results": []})
+    useful = json.dumps({"success": True, "query": "weather", "quality": "high", "results": []})
+
+    assert guard.observe_result("web_search", low_quality) is None
+    assert guard.observe_result("web_search", useful) is None
+    assert guard.observe_result("web_search", low_quality) is None
+    assert guard.observe_result("web_search", low_quality) is None
+
+    assert guard.observe_result("web_search", low_quality) is not None