Continuous read tool calls

Challenges
1

I’m stuck on Stage #FF2

My Prompt: explain what code is written in ./apple.py
apple.py has simple hello world code in python

The first user prompt is sent to the LLM, it responds back with a ReadFile tool call. I return the entire contents of the file in a ToolMessage. Then I expect the LLM to stop and give me the results. But it keeps calling the ReadFile tool.

I’ve tried mentioning explicitly in the tool call description that this returns complete contents of the file in one go. I’m still facing this issue.

Here are my logs:

2026/09/15 02:57:07 INFO Using  model=anthropic/claude-sonnet-4.5
2026/09/15 02:57:12 INFO Chat History value="[\n{\n\"content\": \"explain what code is written in ./apple.py\",\n\"role\": \"user\"\n}\n]"
2026/09/15 02:57:16 INFO Response:  FinishReason=tool_calls
2026/09/15 02:57:16 INFO Tool Call:  "ToolCall count: "=1 FunctionName=ReadFile FunctionArgs="{\"filePath\": \"./apple.py\"}"
2026/09/15 02:57:21 INFO Chat History value="[\n{\n\"content\": \"explain what code is written in ./apple.py\",\n\"role\": \"user\"\n},\n{\n\"content\": \"print(hello world)\\n\",\n\"tool_call_id\": \"toolu_bdrk_01TVFVh6HStYETN2DydB6jEk\",\n\"role\": \"tool\"\n}\n]"
2026/09/15 02:57:23 INFO Response:  FinishReason=tool_calls
2026/09/15 02:57:23 INFO Tool Call:  "ToolCall count: "=1 FunctionName=ReadFile FunctionArgs="{\"filePath\": \"./apple.py\"}"
2026/09/15 02:57:28 INFO Chat History value="[\n{\n\"content\": \"explain what code is written in ./apple.py\",\n\"role\": \"user\"\n},\n{\n\"content\": \"print(hello world)\\n\",\n\"tool_call_id\": \"toolu_bdrk_01TVFVh6HStYETN2DydB6jEk\",\n\"role\": \"tool\"\n},\n{\n\"content\": \"print(hello world)\\n\",\n\"tool_call_id\": \"toolu_bdrk_019NxpLCSaoYUgsfkUhSKcBi\",\n\"role\": \"tool\"\n}\n]"

And here’s a snippet of my code:

	resp, err := client.Chat.Completions.New(context.Background(),
			openai.ChatCompletionNewParams{
				Model: model, 
				Messages: chatHistory,
				Tools: []openai.ChatCompletionToolUnionParam{
					openai.ChatCompletionFunctionTool(
						shared.FunctionDefinitionParam{
							Name:        "ReadFile",
							Description: openai.String("Read File at path, returns complete contents of the file at one go"),
							Parameters: map[string]any{
								"type": "object",
								"properties": map[string]any{
									"filePath": map[string]any{
										"type":        "string",
										"description": "file path to read",
									},
								},
								"required":             []string{"filePath"},
								"additionalProperties": false,
							},
						},
					),
				},
			},
		)
		if err != nil {
			fmt.Fprintf(os.Stderr, "error: %v\n", err)
			os.Exit(1)
		}
		if len(resp.Choices) == 0 {
			panic("No choices in response")
		}

		response := resp.Choices[0]

		slog.Info("Response: ", "FinishReason", response.FinishReason)

		if response.FinishReason == "tool_calls" {
			slog.Info("Tool Call: ", "ToolCall count: ", len(response.Message.ToolCalls), "FunctionName", response.Message.ToolCalls[0].Function.Name, "FunctionArgs", response.Message.ToolCalls[0].Function.Arguments)
			if resp.Choices[0].Message.ToolCalls[0].Function.Name == "ReadFile" {
				var readFileFunctionArguments ReadFileFunctionArguments
				err := UnmarshalJSONString(resp.Choices[0].Message.ToolCalls[0].Function.Arguments, &readFileFunctionArguments)
				if err != nil {
					fmt.Fprintf(os.Stderr, "error: %v\n", err)
					os.Exit(1)
				}
				filePath := readFileFunctionArguments.FilePath
				file, err := os.ReadFile(filePath)
				if err != nil {
					fmt.Fprintf(os.Stderr, "error: %v\n", err)
					os.Exit(1)
				}
				chatHistory = append(chatHistory, openai.ChatCompletionMessageParamUnion{
					OfTool: &openai.ChatCompletionToolMessageParam{
						Content: openai.ChatCompletionToolMessageParamContentUnion{
							OfString: openai.String(string(file)),
						},
						ToolCallID: resp.Choices[0].Message.ToolCalls[0].ID,
					},
				})
			}
		} else if response.FinishReason == "stop" {
			fmt.Println(response.Message.Content)
			break
		}

PS: ignore the optimizations possible, I was hoping to do that after I have a working piece ready.

1 Like