fix deepseek-chat bug (#1066)

Co-authored-by: FantasticCode2019 <1443996278@qq.com>

pkg/agent/context.go

@@ -605,7 +605,60 @@ func sanitizeHistoryForProvider(history []providers.Message) []providers.Message
 		}
 	}
 
-	return sanitized
+	// Second pass: ensure every assistant message with tool_calls has matching
+	// tool result messages following it. This is required by strict providers
+	// like DeepSeek that enforce: "An assistant message with 'tool_calls' must
+	// be followed by tool messages responding to each 'tool_call_id'."
+	final := make([]providers.Message, 0, len(sanitized))
+	for i := 0; i < len(sanitized); i++ {
+		msg := sanitized[i]
+		if msg.Role == "assistant" && len(msg.ToolCalls) > 0 {
+			// Collect expected tool_call IDs
+			expected := make(map[string]bool, len(msg.ToolCalls))
+			for _, tc := range msg.ToolCalls {
+				expected[tc.ID] = false
+			}
+
+			// Check following messages for matching tool results
+			toolMsgCount := 0
+			for j := i + 1; j < len(sanitized); j++ {
+				if sanitized[j].Role != "tool" {
+					break
+				}
+				toolMsgCount++
+				if _, exists := expected[sanitized[j].ToolCallID]; exists {
+					expected[sanitized[j].ToolCallID] = true
+				}
+			}
+
+			// If any tool_call_id is missing, drop this assistant message and its partial tool messages
+			allFound := true
+			for toolCallID, found := range expected {
+				if !found {
+					allFound = false
+					logger.DebugCF(
+						"agent",
+						"Dropping assistant message with incomplete tool results",
+						map[string]any{
+							"missing_tool_call_id": toolCallID,
+							"expected_count":       len(expected),
+							"found_count":          toolMsgCount,
+						},
+					)
+					break
+				}
+			}
+
+			if !allFound {
+				// Skip this assistant message and its tool messages
+				i += toolMsgCount
+				continue
+			}
+		}
+		final = append(final, msg)
+	}
+
+	return final
 }
 
 func (cb *ContextBuilder) AddToolResult(

pkg/agent/context_test.go

@@ -207,3 +207,77 @@ func assertRoles(t *testing.T, msgs []providers.Message, expected ...string) {
 		}
 	}
 }
+
+// TestSanitizeHistoryForProvider_IncompleteToolResults tests the forward validation
+// that ensures assistant messages with tool_calls have ALL matching tool results.
+// This fixes the DeepSeek error: "An assistant message with 'tool_calls' must be
+// followed by tool messages responding to each 'tool_call_id'."
+func TestSanitizeHistoryForProvider_IncompleteToolResults(t *testing.T) {
+	// Assistant expects tool results for both A and B, but only A is present
+	history := []providers.Message{
+		msg("user", "do two things"),
+		assistantWithTools("A", "B"),
+		toolResult("A"),
+		// toolResult("B") is missing - this would cause DeepSeek to fail
+		msg("user", "next question"),
+		msg("assistant", "answer"),
+	}
+
+	result := sanitizeHistoryForProvider(history)
+	// The assistant message with incomplete tool results should be dropped,
+	// along with its partial tool result. The remaining messages are:
+	// user ("do two things"), user ("next question"), assistant ("answer")
+	if len(result) != 3 {
+		t.Fatalf("expected 3 messages, got %d: %+v", len(result), roles(result))
+	}
+	assertRoles(t, result, "user", "user", "assistant")
+}
+
+// TestSanitizeHistoryForProvider_MissingAllToolResults tests the case where
+// an assistant message has tool_calls but no tool results follow at all.
+func TestSanitizeHistoryForProvider_MissingAllToolResults(t *testing.T) {
+	history := []providers.Message{
+		msg("user", "do something"),
+		assistantWithTools("A"),
+		// No tool results at all
+		msg("user", "hello"),
+		msg("assistant", "hi"),
+	}
+
+	result := sanitizeHistoryForProvider(history)
+	// The assistant message with no tool results should be dropped.
+	// Remaining: user ("do something"), user ("hello"), assistant ("hi")
+	if len(result) != 3 {
+		t.Fatalf("expected 3 messages, got %d: %+v", len(result), roles(result))
+	}
+	assertRoles(t, result, "user", "user", "assistant")
+}
+
+// TestSanitizeHistoryForProvider_PartialToolResultsInMiddle tests that
+// incomplete tool results in the middle of a conversation are properly handled.
+func TestSanitizeHistoryForProvider_PartialToolResultsInMiddle(t *testing.T) {
+	history := []providers.Message{
+		msg("user", "first"),
+		assistantWithTools("A"),
+		toolResult("A"),
+		msg("assistant", "done"),
+		msg("user", "second"),
+		assistantWithTools("B", "C"),
+		toolResult("B"),
+		// toolResult("C") is missing
+		msg("user", "third"),
+		assistantWithTools("D"),
+		toolResult("D"),
+		msg("assistant", "all done"),
+	}
+
+	result := sanitizeHistoryForProvider(history)
+	// First round is complete (user, assistant+tools, tool, assistant),
+	// second round is incomplete and dropped (assistant+tools, partial tool),
+	// third round is complete (user, assistant+tools, tool, assistant).
+	// Remaining: user, assistant, tool, assistant, user, user, assistant, tool, assistant
+	if len(result) != 9 {
+		t.Fatalf("expected 9 messages, got %d: %+v", len(result), roles(result))
+	}
+	assertRoles(t, result, "user", "assistant", "tool", "assistant", "user", "user", "assistant", "tool", "assistant")
+}