refactor(tools): use structured tool error results

tests/tools/test_exec_session_tools.py

@@ -11,6 +11,7 @@ from nanobot.agent.tools.exec_session import (
     ListExecSessionsTool,
     WriteStdinTool,
 )
+from nanobot.agent.tools.registry import is_tool_error_result
 from nanobot.agent.tools.shell import ExecTool
 
 
@@ -334,9 +335,10 @@ def test_write_stdin_reports_missing_session(tmp_path):
     manager = ExecSessionManager()
     tool = WriteStdinTool(manager=manager)
 
-    result = asyncio.run(tool.execute(session_id="missing", chars=""))
+    result = asyncio.run(tool.execute(session_id="missing\nExit code: 0", chars=""))
 
-    assert "exec session not found" in result
+    assert result == "Error: exec session not found: 'missing\\nExit code: 0'"
+    assert is_tool_error_result("write_stdin", result)
 
 
 def test_list_exec_sessions_reports_running_commands(tmp_path):

tests/tools/test_tool_registry.py

@@ -1,8 +1,9 @@
 from __future__ import annotations
 
 from typing import Any
+from unittest.mock import AsyncMock
 
-from nanobot.agent.tools.base import Tool
+from nanobot.agent.tools.base import Tool, ToolResult
 from nanobot.agent.tools.filesystem import ReadFileTool
 from nanobot.agent.tools.registry import ToolRegistry
 
@@ -257,6 +258,41 @@ async def test_registry_rejects_unknown_builtin_tool_parameters(tmp_path) -> Non
     assert "one" not in result
 
 
+async def test_registry_preserves_successful_exec_output_that_starts_with_error() -> None:
+    registry = ToolRegistry()
+    output = "Error: generated report successfully\n\nExit code: 0"
+    tool = _FakeTool("exec")
+    tool.execute = AsyncMock(return_value=output)
+    registry.register(tool)
+
+    result = await registry.execute("exec", {})
+
+    assert result == output
+
+
+async def test_registry_uses_structured_tool_result_for_errors() -> None:
+    registry = ToolRegistry()
+    output = "Error: plain tool output, not a structured failure"
+    raw_tool = _FakeTool("raw_output")
+    raw_tool.execute = AsyncMock(return_value=output)
+    registry.register(raw_tool)
+
+    raw_result = await registry.execute("raw_output", {})
+
+    assert raw_result == output
+
+    failing_tool = _FakeTool("failing_tool")
+    failing_tool.execute = AsyncMock(return_value=ToolResult.error("Error: real failure"))
+    registry.register(failing_tool)
+
+    error_result = await registry.execute("failing_tool", {})
+
+    assert isinstance(error_result, ToolResult)
+    assert error_result.is_error
+    assert error_result.startswith("Error: real failure")
+    assert "[Analyze the error above" in error_result
+
+
 def test_get_definitions_returns_cached_result() -> None:
     registry = ToolRegistry()
     registry.register(_FakeTool("read_file"))