Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: stop reading a success line like "Errors: 0" as a failed tool result (#10554)

Nilayb9f1daed00e0
diff --git a/studio/backend/core/inference/tool_call_parser.py b/studio/backend/core/inference/tool_call_parser.py
index a1d36074c68..d18083c51bb 100644
--- a/studio/backend/core/inference/tool_call_parser.py
+++ b/studio/backend/core/inference/tool_call_parser.py
@@ -118,7 +118,8 @@ _TOOL_ALL_PATS = _TOOL_CLOSED_PATS + [
TOOL_ERROR_PREFIXES = (
- "Error",
+ "Error:",
+ "Error ",
"Search failed",
"Execution error",
"Blocked:",
diff --git a/studio/backend/tests/test_tool_loop_controller.py b/studio/backend/tests/test_tool_loop_controller.py
index dcb179980f9..b67dd9fdefe 100644
--- a/studio/backend/tests/test_tool_loop_controller.py
+++ b/studio/backend/tests/test_tool_loop_controller.py
@@ -19,11 +19,12 @@ from core.inference.tool_loop_controller import (
canonical_tool_call_key,
coerce_arguments_by_schema,
coerce_tool_arguments,
+ is_tool_error,
status_for_tool,
strip_result_for_model,
tool_event_provenance,
)
-from core.inference.tool_call_parser import parse_tool_calls_from_text
+from core.inference.tool_call_parser import TOOL_ERROR_NUDGE, parse_tool_calls_from_text
from core.inference.tools import ALL_TOOLS, _mcp_specs_for_server
@@ -444,3 +445,30 @@ def test_a_declared_type_nested_in_a_container_is_read_too():
# An already-typed container is descended into too: its elements can still be text.
call = {"path": "app.py", "edits": json.loads(edits)}
assert coerce_arguments_by_schema(call, props) == {"path": "app.py", "edits": typed}
+
+
+ "result, failed",
+ [
+ ("Error: boom", True),
+ (" Error: boom", True),
+ ("Error executing tool remote_thing: disk full", True),
+ ("Error running command `git push`: permission denied", True),
+ ("Errors: 0", False),
+ ("Errors: none found", False),
+ ("Errored, then recovered", False),
+ ("Error-free run", False),
+ ],
+)
+def test_only_a_delimited_error_marks_a_result_failed(result, failed):
+ assert is_tool_error(result) is failed
+
+
+def test_a_success_that_opens_with_error_is_not_nudged_as_a_failure():
+ controller = ToolLoopController(tools = [_tool("web_search")])
+ decision = controller.prepare_call(_call("web_search", {"url": "https://example.com/log"}))
+
+ completion = controller.record_result(decision, "Errors: 0 across 128 files")
+
+ assert not completion.is_error
+ assert TOOL_ERROR_NUDGE not in completion.model_message()["content"]
diff --git a/studio/backend/tests/test_tool_result_fits_window.py b/studio/backend/tests/test_tool_result_fits_window.py
index 0ce4dcae64c..c497c016af9 100644
--- a/studio/backend/tests/test_tool_result_fits_window.py
+++ b/studio/backend/tests/test_tool_result_fits_window.py
@@ -2095,7 +2095,7 @@ class TestWhatTheLoopAppendsIsPricedToo:
from core.inference.tool_call_parser import TOOL_ERROR_NUDGE
# The same length, so the only thing between them is the nudge one of them will be
- # given: "Error" is a `TOOL_ERROR_PREFIXES` entry and "Alpha" is not.
+ # given: "Error: " opens with a `TOOL_ERROR_PREFIXES` entry and "Alpha: " does not.
failed = self._fitted(monkeypatch, "Error: ")
fine = self._fitted(monkeypatch, "Alpha: ")