cogito/tools_test.go

package cogito_test

import (
	"fmt"
	"strings"

	. "github.com/mudler/cogito"
	"github.com/mudler/cogito/tests/mock"
	. "github.com/onsi/ginkgo/v2"
	. "github.com/onsi/gomega"
	"github.com/sashabaranov/go-openai"
	"github.com/sashabaranov/go-openai/jsonschema"
)

var _ = Describe("ExecuteTools", func() {
	var mockLLM *mock.MockOpenAIClient
	var originalFragment Fragment

	BeforeEach(func() {
		mockLLM = mock.NewMockOpenAIClient()
		originalFragment = NewEmptyFragment().
			AddMessage(UserMessageRole, "What is photosynthesis?").
			AddMessage(AssistantMessageRole, "Photosynthesis is the process by which plants convert sunlight into energy.")
	})

	Context("ToolDefinition", func() {
		It("should create a valid ToolDefinition", func() {
			mockToolDef := mock.NewMockTool("search", "Search for information")
			mockToolDefT := mockToolDef.(*ToolDefinition[map[string]any])
			toolDefinition := ToolDefinition[map[string]any]{
				ToolRunner:  mockToolDefT.ToolRunner,
				Name:        "search",
				Description: "Search for information",
				InputArguments: &struct {
					Query string `json:"query"`
				}{},
			}
			tool := toolDefinition.Tool()
			Expect(tool.Function.Name).To(Equal("search"))
			Expect(tool.Function.Description).To(Equal("Search for information"))
			Expect(tool.Function.Parameters).To(Equal(jsonschema.Definition{
				Type:                 jsonschema.Object,
				AdditionalProperties: false,
				Properties: map[string]jsonschema.Definition{
					"query": {
						Type: jsonschema.String,
						Enum: nil,
					},
				},
				Required: []string{"query"},
				Defs:     map[string]jsonschema.Definition{},
			}))
		})

		It("should create a valid ToolDefinition with enums and description", func() {
			mockToolDef := mock.NewMockTool("search", "Search for information")
			mockToolDefT := mockToolDef.(*ToolDefinition[map[string]any])
			toolDefinition := ToolDefinition[map[string]any]{
				ToolRunner:  mockToolDefT.ToolRunner,
				Name:        "search",
				Description: "Search for information",
				InputArguments: &struct {
					Query string `json:"query" enum:"foo,bar" description:"The query to search for"`
				}{},
			}
			tool := toolDefinition.Tool()
			Expect(tool.Function.Name).To(Equal("search"))
			Expect(tool.Function.Description).To(Equal("Search for information"))
			Expect(tool.Function.Parameters).To(Equal(jsonschema.Definition{
				Type:                 jsonschema.Object,
				AdditionalProperties: false,
				Properties: map[string]jsonschema.Definition{
					"query": {
						Type:        jsonschema.String,
						Enum:        []string{"foo", "bar"},
						Description: "The query to search for",
					},
				},
				Required: []string{"query"},
				Defs:     map[string]jsonschema.Definition{},
			}))
		})

		It("should create a valid ToolDefinition which arg is not required", func() {
			mockToolDef := mock.NewMockTool("search", "Search for information")
			mockToolDefT := mockToolDef.(*ToolDefinition[map[string]any])
			toolDefinition := ToolDefinition[map[string]any]{
				ToolRunner:  mockToolDefT.ToolRunner,
				Name:        "search",
				Description: "Search for information",
				InputArguments: &struct {
					Query string `json:"query" required:"false"`
				}{},
			}
			tool := toolDefinition.Tool()
			Expect(tool.Function.Name).To(Equal("search"))
			Expect(tool.Function.Description).To(Equal("Search for information"))
			Expect(tool.Function.Parameters).To(Equal(jsonschema.Definition{
				Type:                 jsonschema.Object,
				AdditionalProperties: false,
				Properties: map[string]jsonschema.Definition{
					"query": {
						Type: jsonschema.String,
					},
				},
				Required: nil,
				Defs:     map[string]jsonschema.Definition{},
			}))
		})
	})

	Context("ExecuteTools with tools", func() {
		It("should execute tools when provided", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// First tool selection and execution
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "chlorophyll"}`)
			mock.SetRunResult(mockTool, "Chlorophyll is a green pigment found in plants.")
			// After tool execution, ToolReEvaluator (toolSelection) picks next tool
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "grass"}`)

			// Second tool selection and execution
			// (The "grass" tool call above will be picked as nextAction)
			mock.SetRunResult(mockTool, "Grass is a plant that grows on the ground.")
			// After tool execution, ToolReEvaluator (toolSelection) picks next tool
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "baz"}`)

			// Third tool selection and execution
			// (The "baz" tool call above will be picked as nextAction)
			mock.SetRunResult(mockTool, "Baz is a plant that grows on the ground.")

			// After ToolReEvaluator returns no tool, Ask() is called to get final response
			mockLLM.SetAskResponse("Here is the final response with all the information gathered.")

			result, err := ExecuteTools(mockLLM, originalFragment, WithIterations(3), WithTools(mockTool))
			Expect(err).ToNot(HaveOccurred())

			// ExecuteTools now calls Ask() at the end to get a final response
			// when ToolReEvaluator returns no more tools
			Expect(len(mockLLM.FragmentHistory)).To(Equal(1), fmt.Sprintf("Fragment history: %v", mockLLM.FragmentHistory))

			Expect(result).ToNot(BeNil())

			Expect(len(result.Status.ToolsCalled)).To(Equal(3))
			Expect(len(result.Status.ToolResults)).To(Equal(3))

			Expect(result.Status.ToolResults[0].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[0].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[0].Result).To(Equal("Chlorophyll is a green pigment found in plants."))
			Expect(result.Status.ToolResults[1].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[1].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[1].Result).To(Equal("Grass is a plant that grows on the ground."))
			Expect(result.Status.ToolResults[2].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[2].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[2].Result).To(Equal("Baz is a plant that grows on the ground."))
		})

		It("should execute tools when provided with guidelines", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			mockWeatherTool := mock.NewMockTool("get_weather", "Get the weather")
			// First iteration
			// 1. Guidelines selection:
			mockLLM.SetAskResponse("Only the first guideline is relevant.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"guidelines": [1]}`)
			// 2. Tool selection (direct):
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "chlorophyll"}`)
			mock.SetRunResult(mockTool, "Chlorophyll is a green pigment found in plants.")

			// Second iteration
			// 1. Guidelines selection:
			mockLLM.SetAskResponse("Only the first guideline is relevant.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"guidelines": [1]}`)
			// 2. Tool selection:
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "grass"}`)
			mock.SetRunResult(mockTool, "Grass is a plant that grows on the ground.")

			// Third iteration
			// 1. Guidelines selection:
			mockLLM.SetAskResponse("Only the second guideline is relevant.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"guidelines": [2]}`)
			// 2. Tool selection:
			mockLLM.AddCreateChatCompletionFunction("get_weather", `{"query": "baz"}`)
			mock.SetRunResult(mockWeatherTool, "Baz is a plant that grows on the ground.")

			// When max iterations is reached, Ask() is called to get final response
			mockLLM.SetAskResponse("All tasks completed.")

			result, err := ExecuteTools(mockLLM, originalFragment, WithIterations(3), WithTools(mockTool, mockWeatherTool),
				EnableStrictGuidelines,
				WithGuidelines(
					Guideline{
						Condition: "User asks about informations",
						Action:    "Use the search tool to find information.",
						Tools:     Tools{mockTool},
					},
					Guideline{
						Condition: "User asks for the weather in a city",
						Action:    "Use the weather tool to find the weather in the city.",
						Tools:     Tools{mockWeatherTool},
					},
				))
			Expect(err).ToNot(HaveOccurred())

			// Check fragments history to see if we behaved as expected
			// Guidelines selection: 3 iterations × 1 Ask() + Final response when max iterations reached: 1 Ask() = 4 Ask() calls
			Expect(len(mockLLM.FragmentHistory)).To(Equal(4), fmt.Sprintf("Fragment history: %v", mockLLM.FragmentHistory))

			// Iteration 1: [0] Guidelines
			Expect(mockLLM.FragmentHistory[0].String()).To(ContainSubstring("You are an AI assistant that needs to understand if any of the guidelines should be applied"))

			// Iteration 2: [1] Guidelines
			Expect(mockLLM.FragmentHistory[1].String()).To(ContainSubstring("You are an AI assistant that needs to understand if any of the guidelines should be applied"))

			// Iteration 3: [2] Guidelines
			Expect(mockLLM.FragmentHistory[2].String()).To(ContainSubstring("You are an AI assistant that needs to understand if any of the guidelines should be applied"))

			// [3] Final Ask when max iterations reached - check that it contains the conversation
			Expect(mockLLM.FragmentHistory[3].String()).To(ContainSubstring(`get_weather({"query":"baz"})`))
			Expect(result).ToNot(BeNil())

			Expect(len(result.Status.ToolsCalled)).To(Equal(3))
			Expect(len(result.Status.ToolResults)).To(Equal(3))

			Expect(result.Status.ToolResults[0].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[0].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[0].Result).To(Equal("Chlorophyll is a green pigment found in plants."))
			Expect(result.Status.ToolResults[1].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[1].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[1].Result).To(Equal("Grass is a plant that grows on the ground."))
			Expect(result.Status.ToolResults[2].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[2].Name).To(Equal("get_weather"))
			Expect(result.Status.ToolResults[2].Result).To(Equal("Baz is a plant that grows on the ground."))
		})

		It("should execute autoplan basic functionality", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// Mock planning decision - decide that planning is needed
			mockLLM.SetAskResponse("Yes, this task requires planning to be completed effectively.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"extract_boolean": true}`)

			// Mock goal extraction
			mockLLM.SetAskResponse("The goal is to research information about photosynthesis.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"goal": "Research information about photosynthesis"}`)

			// Mock plan creation (first step of plan extraction)
			mockLLM.SetAskResponse("Here is a plan with subtasks: 1. Search for basic information about photosynthesis")

			// Mock subtask extraction (second step of plan extraction) - this uses CreateChatCompletion
			mockLLM.AddCreateChatCompletionFunction("json", `{"subtasks": ["Search for basic information about photosynthesis"]}`)

			// Mock first subtask execution - search
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "photosynthesis basics"}`)
			mockLLM.SetAskResponse("Photosynthesis is the process by which plants convert sunlight into energy.")
			mock.SetRunResult(mockTool, "Photosynthesis is the process by which plants convert sunlight into energy.")
			mockLLM.SetAskResponse("Photosynthesis is the process by which plants convert sunlight into energy.")

			// Mock goal achievement check for first subtask
			mockLLM.AddCreateChatCompletionFunction("json", `{"extract_boolean": true}`)

			result, err := ExecuteTools(mockLLM, originalFragment,
				EnableAutoPlan,
				WithTools(mockTool))

			Expect(err).ToNot(HaveOccurred())
			Expect(result).ToNot(BeNil())

			// Verify that planning was executed by checking fragment history
			// PlanDecision + GoalExtraction + PlanCreation + GoalCheck = 4 Ask() calls
			// ToolReEvaluator uses toolSelection (CreateChatCompletion), not Ask()
			Expect(len(mockLLM.FragmentHistory)).To(BeNumerically("==", 5), fmt.Sprintf("Fragment history: %v", mockLLM.FragmentHistory))

			// Check that planning decision was made
			Expect(mockLLM.FragmentHistory[0].String()).To(
				And(
					ContainSubstring("You are an AI assistant that decides if planning and executing subtasks in sequence is needed from a conversation"),
					ContainSubstring("What is photosynthesis"),
				))

			// Check that goal extraction was called
			Expect(mockLLM.FragmentHistory[1].String()).To(
				ContainSubstring("Analyze the following text and the context to identify the goal"))

			// Check that plan creation was called
			Expect(mockLLM.FragmentHistory[2].String()).To(
				ContainSubstring("You are an AI assistant that breaks down a goal into a series of actionable steps"))

			Expect(len(result.Messages)).To(Equal(4), fmt.Sprintf("Messages: %+v", result.Messages))

			Expect(result.Messages[len(result.Messages)-1].Content).To(
				And(
					ContainSubstring("hotosynthesis is the process by which plants convert sunlight into energy"),
				),
				fmt.Sprintf("Result: %+v", result),
			)

			Expect(len(result.Status.Plans)).To(Equal(1))

			// Verify tools were called correctly
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			Expect(len(result.Status.ToolResults)).To(Equal(1))

			Expect(result.Status.ToolResults[0].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[0].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[0].Result).To(Equal("Photosynthesis is the process by which plants convert sunlight into energy."))
		})

		It("should not execute autoplan when planning is not needed", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// Mock planning decision - decide that planning is NOT needed
			mockLLM.SetAskResponse("No, this task does not require planning.")
			mockLLM.AddCreateChatCompletionFunction("json", `{"extract_boolean": false}`)

			// Mock regular tool execution (since planning is not needed, it falls back to normal tool execution)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "photosynthesis"}`)
			mockLLM.SetAskResponse("Photosynthesis is the process by which plants convert sunlight into energy.")
			mock.SetRunResult(mockTool, "Photosynthesis is the process by which plants convert sunlight into energy.")
			// After tool execution, ToolReEvaluator (toolSelection) returns no tool (text response)
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment,
				EnableAutoPlan,
				WithTools(mockTool))

			Expect(err).ToNot(HaveOccurred())
			Expect(result).ToNot(BeNil())

			// Verify that planning decision was made but no plan was executed
			// PlanDecision = 1 Ask() call
			// ToolReEvaluator uses toolSelection (CreateChatCompletion), not Ask()
			Expect(len(mockLLM.FragmentHistory)).To(Equal(2), fmt.Sprintf("Fragment history: %v", mockLLM.FragmentHistory))

			// Check that planning decision was made
			Expect(mockLLM.FragmentHistory[0].String()).To(
				And(
					ContainSubstring("You are an AI assistant that decides if planning and executing subtasks in sequence is needed from a conversation"),
					ContainSubstring("What is photosynthesis"),
				))

			// Check that tools were called (regular tool execution, not planning)
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			Expect(len(result.Status.ToolResults)).To(Equal(1))

			Expect(result.Status.ToolResults[0].Executed).To(BeTrue())
			Expect(result.Status.ToolResults[0].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[0].Result).To(Equal("Photosynthesis is the process by which plants convert sunlight into energy."))
		})

	})

	Context("Tool Call Callbacks", func() {
		It("should call the callback with ToolChoice and SessionState", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			var receivedTool *ToolChoice
			var receivedState *SessionState

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Test result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					receivedTool = tool
					receivedState = state
					return ToolCallDecision{Approved: true}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(receivedTool).ToNot(BeNil())
			Expect(receivedTool.Name).To(Equal("search"))
			Expect(receivedState).ToNot(BeNil())
			Expect(receivedState.ToolChoice).To(Equal(receivedTool))
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
		})

		It("should interrupt execution when Approved is false", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					return ToolCallDecision{Approved: false}
				}))

			Expect(err).To(HaveOccurred())
			Expect(err).To(Equal(ErrToolCallCallbackInterrupted))
			Expect(len(result.Status.ToolsCalled)).To(Equal(0))
		})

		It("should skip tool call when Skip is true", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			// After skipping, ToolReEvaluator returns no tool (this happens after the skip)
			mockLLM.SetAskResponse("LLM result")
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					return ToolCallDecision{Approved: true, Skip: true}
				}))

			// When skipping with DisableToolReEvaluator, we might get ErrNoToolSelected
			// because no tools were actually executed. This is expected behavior.
			if err != nil {
				Expect(err).To(Equal(ErrNoToolSelected))
			}
			// Tool should not be executed
			Expect(len(result.Status.ToolsCalled)).To(Equal(0))
			// But should be in the conversation
			Expect(len(result.Messages)).To(BeNumerically(">", 0))
			// Check that skip message was added
			foundSkipMessage := false
			for _, msg := range result.Messages {
				if msg.Role == "tool" && msg.Content == "Tool call skipped by user" {
					foundSkipMessage = true
					break
				}
			}
			Expect(foundSkipMessage).To(BeTrue())
		})

		It("should use directly modified tool choice when Modified is set", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// First tool selection (will be modified)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			mock.SetRunResult(mockTool, "Modified result")
			mockLLM.SetAskResponse("LLM result")

			var executedArgs map[string]any
			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					// Directly modify the tool arguments
					modified := *tool
					modified.Arguments = map[string]any{
						"query": "modified_query",
					}
					return ToolCallDecision{
						Approved: true,
						Modified: &modified,
					}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			Expect(result.LastMessage().Content).To(Equal("LLM result"))
			// Check that the modified arguments were used
			executedArgs = result.Status.ToolResults[0].ToolArguments.Arguments
			Expect(executedArgs["query"]).To(Equal("modified_query"))
		})

		It("should handle adjustment feedback and re-evaluate tool call", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			callbackCount := 0

			// First tool selection (original)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			// Adjustment: LLM re-evaluates with feedback
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted"}`)
			mock.SetRunResult(mockTool, "Adjusted result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithMaxAdjustmentAttempts(3),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					callbackCount++
					if callbackCount == 1 {
						// First call: provide adjustment feedback
						return ToolCallDecision{
							Approved:   true,
							Adjustment: "Please use a more specific query",
						}
					}
					// Second call: approve the adjusted tool
					return ToolCallDecision{Approved: true}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(callbackCount).To(Equal(2)) // Called twice: original + adjusted
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			// Check that adjusted arguments were used
			Expect(result.Status.ToolResults[0].ToolArguments.Arguments["query"]).To(Equal("adjusted"))
		})

		It("should respect max adjustment attempts limit", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			callbackCount := 0

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			// Adjustment attempts (will hit max)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted1"}`)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted2"}`)
			mock.SetRunResult(mockTool, "Final result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithMaxAdjustmentAttempts(2), // Limit to 2 attempts
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					callbackCount++
					// Always provide adjustment to test max limit
					if callbackCount < 3 {
						return ToolCallDecision{
							Approved:   true,
							Adjustment: "Keep adjusting",
						}
					}
					return ToolCallDecision{Approved: true}
				}))

			Expect(err).ToNot(HaveOccurred())
			// Should have hit max attempts (2 adjustments + 1 final = 3 calls max)
			Expect(callbackCount).To(BeNumerically("<=", 3))
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
		})

		It("should handle skip during adjustment loop", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			callbackCount := 0

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			// Adjustment attempt
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted"}`)
			mockLLM.SetAskResponse("LLM result")

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					callbackCount++
					if callbackCount == 1 {
						// First call: provide adjustment
						return ToolCallDecision{
							Approved:   true,
							Adjustment: "Please adjust",
						}
					}
					// Second call: skip
					return ToolCallDecision{
						Approved: true,
						Skip:     true,
					}
				}))

			// When skipping with DisableToolReEvaluator, we might get ErrNoToolSelected
			if err != nil {
				Expect(err).To(Equal(ErrNoToolSelected))
			}
			Expect(callbackCount).To(Equal(2))
			// Tool should not be executed
			Expect(len(result.Status.ToolsCalled)).To(Equal(0))
		})

		It("should handle direct modification during adjustment loop", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			mock.SetRunResult(mockTool, "Directly modified result")

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			// Adjustment attempt (will be modified directly, so this won't be used)
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted"}`)
			mockLLM.SetAskResponse("LLM result")
			// After modification, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					// First call: provide adjustment
					if tool.Arguments["query"] == "original" {
						return ToolCallDecision{
							Approved:   true,
							Adjustment: "Please adjust",
						}
					}
					// During adjustment: directly modify
					modified := *tool
					modified.Arguments = map[string]any{
						"query": "directly_modified",
					}
					return ToolCallDecision{
						Approved: true,
						Modified: &modified,
					}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			// Check that directly modified arguments were used
			Expect(result.Status.ToolResults[0].ToolArguments.Arguments["query"]).To(Equal("directly_modified"))
		})
	})

	Context("SessionState and Resume", func() {
		It("should create SessionState with ToolChoice and Fragment", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			var savedState *SessionState

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Test result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			_, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					savedState = state
					return ToolCallDecision{Approved: true}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(savedState).ToNot(BeNil())
			Expect(savedState.ToolChoice).ToNot(BeNil())
			Expect(savedState.ToolChoice.Name).To(Equal("search"))
			Expect(savedState.Fragment).ToNot(BeNil())
		})

		It("should resume execution from SessionState", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			var savedState *SessionState

			// First execution - interrupt after saving state
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mockLLM.SetAskResponse("LLM result")
			_, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					savedState = state
					return ToolCallDecision{Approved: false} // Interrupt
				}))

			Expect(err).To(HaveOccurred())
			Expect(savedState).ToNot(BeNil())

			// Resume execution
			mock.SetRunResult(mockTool, "Resumed result")
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			resumedFragment, err := savedState.Resume(mockLLM, WithTools(mockTool))
			Expect(err).ToNot(HaveOccurred())
			Expect(len(resumedFragment.Status.ToolsCalled)).To(Equal(1))
			Expect(resumedFragment.Status.ToolResults[0].Result).To(Equal("Resumed result"))
		})
	})

	Context("WithStartWithAction", func() {
		It("should start execution with a pre-selected tool", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			mock.SetRunResult(mockTool, "Pre-selected result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			initialTool := &ToolChoice{
				Name: "search",
				Arguments: map[string]any{
					"query": "pre_selected_query",
				},
			}

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithStartWithAction(initialTool))

			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
			Expect(result.Status.ToolResults[0].ToolArguments.Arguments["query"]).To(Equal("pre_selected_query"))
		})

		It("should start execution with multiple pre-selected tools", func() {
			mockSearchTool := mock.NewMockTool("search", "Search for information")
			mockWeatherTool := mock.NewMockTool("get_weather", "Get weather information")
			mock.SetRunResult(mockSearchTool, "Search result")
			mock.SetRunResult(mockWeatherTool, "Weather result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			initialTools := []*ToolChoice{
				{
					Name: "search",
					Arguments: map[string]any{
						"query": "test query",
					},
				},
				{
					Name: "get_weather",
					Arguments: map[string]any{
						"city": "San Francisco",
					},
				},
			}

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockSearchTool, mockWeatherTool),
				WithStartWithAction(initialTools...))

			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(2))
			Expect(result.Status.ToolResults[0].ToolArguments.Arguments["query"]).To(Equal("test query"))
			Expect(result.Status.ToolResults[1].ToolArguments.Arguments["city"]).To(Equal("San Francisco"))
		})
	})

	Context("Multiple Tool Selection", func() {
		It("should handle multiple tool selections sequentially", func() {
			mockSearchTool := mock.NewMockTool("search", "Search for information")
			mockWeatherTool := mock.NewMockTool("get_weather", "Get weather information")
			mock.SetRunResult(mockSearchTool, "Search result")
			mock.SetRunResult(mockWeatherTool, "Weather result")
			mockLLM.SetAskResponse("LLM result")

			// LLM selects multiple tools in a single response
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role: AssistantMessageRole.String(),
							ToolCalls: []openai.ToolCall{
								{
									ID:   "call_1",
									Type: openai.ToolTypeFunction,
									Function: openai.FunctionCall{
										Name:      "search",
										Arguments: `{"query": "test"}`,
									},
								},
								{
									ID:   "call_2",
									Type: openai.ToolTypeFunction,
									Function: openai.FunctionCall{
										Name:      "get_weather",
										Arguments: `{"city": "SF"}`,
									},
								},
							},
						},
					},
				},
			})

			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockSearchTool, mockWeatherTool))
			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(2))
			Expect(result.Status.ToolResults[0].Name).To(Equal("search"))
			Expect(result.Status.ToolResults[1].Name).To(Equal("get_weather"))
		})
	})

	Context("Parallel Tool Execution", func() {
		It("should execute multiple tools in parallel when enabled", func() {
			mockSearchTool := mock.NewMockTool("search", "Search for information")
			mockWeatherTool := mock.NewMockTool("get_weather", "Get weather information")
			mock.SetRunResult(mockSearchTool, "Search result")
			mock.SetRunResult(mockWeatherTool, "Weather result")
			mockLLM.SetAskResponse("LLM result")
			// LLM selects multiple tools using the parallel intention tool
			// First, reasoning step - now uses the reasoning tool
			mockLLM.AddCreateChatCompletionFunction("reasoning", `{"reasoning": "I need to search and get weather information."}`)

			// Then, tool selection with multiple tools
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role: AssistantMessageRole.String(),
							ToolCalls: []openai.ToolCall{
								{
									ID:   "call_1",
									Type: openai.ToolTypeFunction,
									Function: openai.FunctionCall{
										Name:      "pick_tools",
										Arguments: `{"tools": ["search", "get_weather"], "reasoning": "Both tools are needed"}`,
									},
								},
							},
						},
					},
				},
			})

			// Parameter generation for search (with forced reasoning, needs parameter reasoning first)
			// 1. Parameter reasoning for search - now uses the reasoning tool
			mockLLM.AddCreateChatCompletionFunction("reasoning", `{"reasoning": "The search tool needs a query parameter to search for information."}`)
			// 2. Parameter generation for search
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			// Parameter generation for weather (with forced reasoning, needs parameter reasoning first)
			// 3. Parameter reasoning for weather - now uses the reasoning tool
			mockLLM.AddCreateChatCompletionFunction("reasoning", `{"reasoning": "The weather tool needs a city parameter to get weather information."}`)
			// 4. Parameter generation for weather
			mockLLM.AddCreateChatCompletionFunction("get_weather", `{"city": "SF"}`)

			// After tool execution, ToolReEvaluator uses forced reasoning, so it needs:
			// 1. Reasoning step response - now uses the reasoning tool
			mockLLM.AddCreateChatCompletionFunction("reasoning", `{"reasoning": "The tools have been executed successfully. No more tools are needed."}`)
			// 2. Intention tool response (return sink state "reply" to indicate no tools needed)
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role: AssistantMessageRole.String(),
							ToolCalls: []openai.ToolCall{
								{
									ID:   "call_reval",
									Type: openai.ToolTypeFunction,
									Function: openai.FunctionCall{
										Name:      "pick_tools",
										Arguments: `{"tools": ["reply"], "reasoning": "No more tools needed"}`,
									},
								},
							},
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment,
				WithTools(mockSearchTool, mockWeatherTool),
				EnableParallelToolExecution,
				WithForceReasoning(),
				WithForceReasoningTool())
			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Status.ToolsCalled)).To(Equal(2))
		})
	})

	Context("WithMaxAdjustmentAttempts", func() {
		It("should use default max adjustment attempts when not specified", func() {
			mockTool := mock.NewMockTool("search", "Search for information")
			callbackCount := 0

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "original"}`)
			// Adjustment attempts
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "adjusted"}`)
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")
			// After tool execution, ToolReEvaluator returns no tool
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				// Don't specify WithMaxAdjustmentAttempts - should use default (5)
				WithToolCallBack(func(tool *ToolChoice, state *SessionState) ToolCallDecision {
					callbackCount++
					if callbackCount == 1 {
						return ToolCallDecision{
							Approved:   true,
							Adjustment: "Adjust",
						}
					}
					return ToolCallDecision{Approved: true}
				}))

			Expect(err).ToNot(HaveOccurred())
			Expect(callbackCount).To(Equal(2))
			Expect(len(result.Status.ToolsCalled)).To(Equal(1))
		})
	})
})

var _ = Describe("ExecuteTools with Compaction", func() {
	var mockLLM *mock.MockOpenAIClient
	var originalFragment Fragment

	BeforeEach(func() {
		mockLLM = mock.NewMockOpenAIClient()
		originalFragment = NewEmptyFragment().
			AddMessage(UserMessageRole, "Task 1").
			AddMessage(AssistantMessageRole, "Done 1")
	})

	Context("WithCompactionThreshold", func() {
		It("should not compact when threshold is disabled (0)", func() {
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)

			mockTool := mock.NewMockTool("search", "Search for information")
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")
			mockLLM.SetUsage(100, 100, 1000)
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Usage: openai.Usage{
					TotalTokens:      1000,
					PromptTokens:     100,
					CompletionTokens: 100,
				},
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			result, err := ExecuteTools(mockLLM, originalFragment, WithTools(mockTool),
				WithCompactionThreshold(0),
			)

			Expect(err).ToNot(HaveOccurred())
			Expect(len(result.Messages)).ToNot(Equal(len(originalFragment.Messages)), fmt.Sprintf("result: %+v", result))
			Expect(result.Status.LastUsage.TotalTokens).To(BeNumerically(">", 0))
			Expect(len(result.Messages)).To(Equal(6))
		})

		It("should not compact when tokens below threshold", func() {
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mockTool := mock.NewMockTool("search", "Search for information")
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			// Create fragment with low token count
			smallFragment := NewEmptyFragment().
				AddMessage(UserMessageRole, "Hi").
				AddMessage(AssistantMessageRole, "Hello")

			result, err := ExecuteTools(mockLLM, smallFragment, WithTools(mockTool),
				WithCompactionThreshold(100000),
				WithCompactionKeepMessages(2))

			Expect(err).ToNot(HaveOccurred())
			// Should not be compacted - still has original messages
			Expect(len(result.Messages)).To(BeNumerically(">", 2))
		})

		It("should compact when token threshold is exceeded", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			// First tool selection
			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")

			// After tool execution, no more tools needed
			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			// Create a large fragment with high token count
			largeFragment := NewEmptyFragment().
				AddMessage(UserMessageRole, "Task 1").
				AddMessage(AssistantMessageRole, "Answer to task 1").
				AddMessage(ToolMessageRole, "Result 1").
				AddMessage(UserMessageRole, "Task 2").
				AddMessage(AssistantMessageRole, "Answer to task 2").
				AddMessage(ToolMessageRole, "Result 2")

			// Set the usage to exceed threshold
			mockLLM.SetUsage(100, 100, 5000)

			// Mock the compaction summary response
			summaryFragment := NewEmptyFragment().
				AddMessage(AssistantMessageRole, "Summary of conversation history.")
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)

			result, err := ExecuteTools(mockLLM, largeFragment, WithTools(mockTool),
				WithCompactionThreshold(1000),
				WithCompactionKeepMessages(1))

			Expect(err).ToNot(HaveOccurred())

			Expect(len(result.Messages)).To(BeNumerically(">", 0))

			Expect(result.Messages[0].Role).To(Equal("system"), fmt.Sprintf("result: %+v", result))
			Expect(result.Messages[0].Content).To(ContainSubstring("compacted"), fmt.Sprintf("result: %+v", result))
			Expect(len(result.Messages)).To(BeNumerically("<", len(largeFragment.Messages)))
		})

		It("should preserve parent fragment after compaction", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")

			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			// Create a fragment with a parent
			parentFragment := NewEmptyFragment().AddMessage(UserMessageRole, "Parent task")
			largeFragment := NewEmptyFragment().
				AddMessage(UserMessageRole, "Task 1").
				AddMessage(AssistantMessageRole, strings.Repeat("response ", 5000))
			largeFragment.ParentFragment = &parentFragment

			// Set usage to exceed threshold
			mockLLM.SetUsage(100, 100, 5000)

			// Mock the compaction summary response (may be used in-loop and again before final Ask)
			summaryFragment := NewEmptyFragment().
				AddMessage(AssistantMessageRole, "Summary of conversation.")
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)

			result, err := ExecuteTools(mockLLM, largeFragment, WithTools(mockTool),
				WithCompactionThreshold(1000),
				WithCompactionKeepMessages(1))

			Expect(err).ToNot(HaveOccurred())
			Expect(result.ParentFragment).ToNot(BeNil())
			Expect(result.ParentFragment.Messages[0].Role).To(Equal(UserMessageRole.String()))
		})

		It("should preserve status after compaction", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")

			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			// Create fragment with status
			largeFragment := NewEmptyFragment().
				AddMessage(UserMessageRole, "Task 1").
				AddMessage(AssistantMessageRole, strings.Repeat("response ", 5000))
			largeFragment.Status = &Status{
				Iterations:   5,
				ReasoningLog: []string{"reasoning1", "reasoning2"},
			}

			// Set usage to exceed threshold
			mockLLM.SetUsage(100, 100, 5000)

			// Mock the compaction summary response (may be used in-loop and again before final Ask)
			summaryFragment := NewEmptyFragment().
				AddMessage(AssistantMessageRole, "Summary of conversation.")
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)

			result, err := ExecuteTools(mockLLM, largeFragment, WithTools(mockTool),
				WithCompactionThreshold(1000),
				WithCompactionKeepMessages(1))

			Expect(err).ToNot(HaveOccurred())
			Expect(result.Status).ToNot(BeNil())
			// Original had Iterations: 5; one tool loop iteration was run, so 6
			Expect(result.Status.Iterations).To(Equal(6))
			Expect(result.Status.ReasoningLog).To(Equal([]string{"reasoning1", "reasoning2"}))
		})

		It("should use rough token estimate when LastUsage is not set", func() {
			mockTool := mock.NewMockTool("search", "Search for information")

			mockLLM.AddCreateChatCompletionFunction("search", `{"query": "test"}`)
			mock.SetRunResult(mockTool, "Result")
			mockLLM.SetAskResponse("LLM result")

			mockLLM.SetCreateChatCompletionResponse(openai.ChatCompletionResponse{
				Choices: []openai.ChatCompletionChoice{
					{
						Message: openai.ChatCompletionMessage{
							Role:    AssistantMessageRole.String(),
							Content: "No more tools needed.",
						},
					},
				},
			})

			// Large fragment without LastUsage set
			largeFragment := NewEmptyFragment().
				AddMessage(UserMessageRole, "Task 1").
				AddMessage(AssistantMessageRole, strings.Repeat("response with lots of content ", 500)).
				AddMessage(ToolMessageRole, "Result 1")

			// Mock the compaction summary response (may be used in-loop and again before final Ask)
			summaryFragment := NewEmptyFragment().
				AddMessage(AssistantMessageRole, "Summary.")
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)
			mockLLM.AskResponses = append([]Fragment{summaryFragment}, mockLLM.AskResponses...)

			result, err := ExecuteTools(mockLLM, largeFragment, WithTools(mockTool),
				WithCompactionThreshold(1000),
				WithCompactionKeepMessages(1))

			Expect(err).ToNot(HaveOccurred())
			// Should be compacted based on rough estimate
			if len(result.Messages) > 0 {
				Expect(result.Messages[0].Role).To(Equal("system"))
			}
		})
	})
})