Skip to content

Personal AI, On Personal Devices

Python

fix(spec-search): use Message objects in run_self_on_task and planner prompt fix (#556)

Stéphane Gazaille71cf1327a5d7
diff --git a/src/openjarvis/learning/spec_search/diagnose/tools.py b/src/openjarvis/learning/spec_search/diagnose/tools.py
index e7c013b3..c829e283 100644
--- a/src/openjarvis/learning/spec_search/diagnose/tools.py
+++ b/src/openjarvis/learning/spec_search/diagnose/tools.py
@@ -16,6 +16,7 @@ from __future__ import annotations
import json
from typing import Any
+from openjarvis.core.types import Message, Role
from openjarvis.learning.spec_search.diagnose.types import DiagnosticTool
@@ -278,7 +279,7 @@ def build_diagnostic_tools(
if sample is None:
return json.dumps({"error": f"Task {task_id} not found in benchmark"})
response = teacher_engine.generate(
- messages=[{"role": "user", "content": sample.query}],
+ messages=[Message(role=Role.USER, content=sample.query)],
model=teacher_model,
max_tokens=max_tokens,
)
diff --git a/src/openjarvis/learning/spec_search/plan/planner.py b/src/openjarvis/learning/spec_search/plan/planner.py
index 0fb106ed..9f0ebf90 100644
--- a/src/openjarvis/learning/spec_search/plan/planner.py
+++ b/src/openjarvis/learning/spec_search/plan/planner.py
@@ -45,7 +45,8 @@ INTELLIGENCE pillar:
AGENT pillar:
- replace_system_prompt: {{"new_content": "You are a helpful assistant.\\n..."}}
-- patch_system_prompt: {{"diff": "--- a/prompt.md\\n+++ b/prompt.md\\n@@ ...\\n"}}
+- patch_system_prompt: {{"diff": "--- a/prompt.md\\n+++ b/prompt.md\\n\
+@@ -3,1 +3,1 @@\\n-Old line\\n+New line\\n"}}
- set_agent_class: {{"agent": "simple", "new_class": "react"}}
- set_agent_param: {{"agent": "native_react", "param": "max_turns", "value": 10}}
- edit_few_shot_exemplars: {{"agent": "native_react", \
diff --git a/tests/learning/spec_search/test_diagnose_tools.py b/tests/learning/spec_search/test_diagnose_tools.py
index 63b208ae..a59fa92c 100644
--- a/tests/learning/spec_search/test_diagnose_tools.py
+++ b/tests/learning/spec_search/test_diagnose_tools.py
@@ -280,3 +280,51 @@ class TestListPersonalBenchmark:
parsed = json.loads(result)
assert len(parsed) == 1
assert parsed[0]["task_id"] == "task-001"
+
+
+class TestRunSelfOnTask:
+ """Tests for the run_self_on_task diagnostic tool."""
+
+ def test_passes_message_objects_to_generate(self, tmp_path: Path) -> None:
+ from openjarvis.core.types import Message, Role
+ from openjarvis.learning.spec_search.diagnose.tools import (
+ build_diagnostic_tools,
+ )
+
+ captured_messages: list[Message] = []
+
+ def _mock_generate(*, messages: list[Message], **_: Any) -> dict[str, Any]:
+ captured_messages.extend(messages)
+ return {
+ "content": "ok",
+ "cost_usd": 0.0,
+ "usage": {"total_tokens": 1},
+ }
+
+ teacher_engine = MagicMock()
+ teacher_engine.generate.side_effect = _mock_generate
+
+ tools = build_diagnostic_tools(
+ trace_store=_make_stub_trace_store(),
+ config=_make_stub_config(tmp_path),
+ benchmark_samples=_make_stub_benchmark_samples(),
+ student_runner=MagicMock(),
+ teacher_engine=teacher_engine,
+ teacher_model="claude-opus-4-6",
+ judge=MagicMock(),
+ session_id="session-001",
+ )
+ run_self_on_task = next(t for t in tools if t.name == "run_self_on_task")
+ result = run_self_on_task.fn(task_id="task-001")
+
+ teacher_engine.generate.assert_called_once()
+ assert captured_messages
+ assert isinstance(captured_messages[0], Message)
+ assert captured_messages[0].role == Role.USER
+
+ parsed = json.loads(result)
+ assert parsed["task_id"] == "task-001"
+ assert parsed["output"] == "ok"
+ assert parsed["cost_usd"] == 0.0
+ assert parsed["tokens_used"] == 1
+ assert "error" not in parsed