fix: wire up enabled_tools in agent loop + simplify sandbox tool selection

Completes the fix started in 8318a51 — handle_function_call() accepted
enabled_tools but run_agent.py never passed it. Now both call sites in
_execute_tool_calls() pass self.valid_tool_names, so each agent session
uses its own tool list instead of the process-global
_last_resolved_tool_names (which subagents can overwrite).

Also simplifies the redundant ternary in code_execution_tool.py:
sandbox_tools is already computed correctly (intersection with session
tools, or full SANDBOX_ALLOWED_TOOLS as fallback), so the conditional
was dead logic.

Inspired by PR #663 (JasonOA888). Closes #662.
Tests: 2857 passed.
This commit is contained in:
teknium1 2026-03-10 06:35:28 -07:00
parent a2ea85924a
commit 771969f747
3 changed files with 18 additions and 7 deletions

View file

@ -601,7 +601,10 @@ class TestExecuteToolCalls:
messages = []
with patch("run_agent.handle_function_call", return_value="search result") as mock_hfc:
agent._execute_tool_calls(mock_msg, messages, "task-1")
mock_hfc.assert_called_once_with("web_search", {"q": "test"}, "task-1")
# enabled_tools passes the agent's own valid_tool_names
args, kwargs = mock_hfc.call_args
assert args[:3] == ("web_search", {"q": "test"}, "task-1")
assert set(kwargs.get("enabled_tools", [])) == agent.valid_tool_names
assert len(messages) == 1
assert messages[0]["role"] == "tool"
assert "search result" in messages[0]["content"]
@ -627,7 +630,9 @@ class TestExecuteToolCalls:
with patch("run_agent.handle_function_call", return_value="ok") as mock_hfc:
agent._execute_tool_calls(mock_msg, messages, "task-1")
# Invalid JSON args should fall back to empty dict
mock_hfc.assert_called_once_with("web_search", {}, "task-1")
args, kwargs = mock_hfc.call_args
assert args[:3] == ("web_search", {}, "task-1")
assert set(kwargs.get("enabled_tools", [])) == agent.valid_tool_names
assert len(messages) == 1
assert messages[0]["role"] == "tool"
assert messages[0]["tool_call_id"] == "c1"