Merge remote-tracking branch 'sandbox/194/work' into feature/ctrlc-cancel

src/harness.cpp

@@ -161,6 +161,12 @@ std::expected<std::string, std::string> Harness::chat(const Message& user_messag
                 tool_result.content = result;
                 push_message(tool_result);
             }
+
+            if (ctrlc_request)
+                // Cancellation after tool execution: same semantics as above -
+                // return empty string to preserve conversation state in the chat file.
+                return "";
+
             continue;
         }
 

tests/CMakeLists.txt

@@ -20,6 +20,7 @@ add_dr_test(test_tool_list_files)
 add_dr_test(test_tool_grep)
 add_dr_test(test_tool_shell)
 add_dr_test(test_tool_shell_ctrlc)
+add_dr_test(test_ctrlc_after_tool)
 add_dr_test(test_tool_read_multi_modal)
 
 # Add feature validation tests

tests/test_ctrlc_after_tool.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# test_ctrlc_after_tool.sh - Test ctrlc (SIGINT) handling between tool calls
+
+# Create mock responses: two tool calls followed by a final response
+# First tool sleeps 1 second, second tool is instant
+# We send SIGINT during the first tool's execution (at 0.5s)
+# The new ctrlc check after tool execution should catch it at 1s
+tool_call run_dangerous_shell '{"command":"sleep 1","timeout_seconds":5}' >> mock_ctrlc_between_tools.txt
+tool_call run_dangerous_shell '{"command":"echo second_tool","timeout_seconds":5}' >> mock_ctrlc_between_tools.txt
+be_done 'Final response' >> mock_ctrlc_between_tools.txt
+
+# Run drharness in background
+"${DRHARNESS}" -F chat_ctrlc.json -n -m "Test ctrlc between tools" -t run_dangerous_shell -d < "mock_ctrlc_between_tools.txt" > /dev/null 2>&1 &
+DRHARNESS_PID=$!
+
+# Send SIGINT at 0.5s - during first tool's sleep
+sleep 0.5
+kill -SIGINT $DRHARNESS_PID 2>/dev/null || true
+
+# Wait for drharness to complete (first tool finishes at 1s, then ctrlc check catches it)
+wait $DRHARNESS_PID 2>/dev/null || true
+
+# Verify the chat file exists and has the conversation state preserved
+if [ ! -f chat_ctrlc.json ]; then
+    echo "FAIL: chat_ctrlc.json was not created"
+    exit 1
+fi
+
+# Check that we have exactly 3 history entries: user message, assistant tool call, tool result
+# The second tool call should NOT have executed because ctrlc was caught after first tool
+HISTORY_LENGTH=$(history_length chat_ctrlc.json)
+
+if [ "$HISTORY_LENGTH" -ne 3 ]; then
+    echo "FAIL: Expected exactly 3 history entries (user, assistant, tool), got $HISTORY_LENGTH"
+    echo "This indicates the second tool call was not cancelled properly"
+    cat chat_ctrlc.json
+    exit 1
+fi
+
+# Verify user message is present
+USER_ROLE=$(dr_role chat_ctrlc.json 0)
+if [ "$USER_ROLE" != "user" ]; then
+    echo "FAIL: Expected user role at index 0, got $USER_ROLE"
+    exit 1
+fi
+
+# Verify assistant with tool call is present
+ASSISTANT_ROLE=$(dr_role chat_ctrlc.json 1)
+if [ "$ASSISTANT_ROLE" != "assistant" ]; then
+    echo "FAIL: Expected assistant role at index 1, got $ASSISTANT_ROLE"
+    exit 1
+fi
+
+# Verify tool result is present
+TOOL_ROLE=$(dr_role chat_ctrlc.json 2)
+if [ "$TOOL_ROLE" != "tool" ]; then
+    echo "FAIL: Expected tool role at index 2, got $TOOL_ROLE"
+    exit 1
+fi
+
+echo "PASS: ctrlc_after_tool"