feat: Implement TodoWrite LLM tool for structured task lists (#58)

Add TodoWrite tool that enables LLMs to create and manage structured
task lists for coding sessions.

## Features
- JSON Schema validation with required fields (id, content, status)
- Single in_progress task invariant enforcement
- Duplicate ID detection and validation
- Replace-all semantics for todo list updates
- Integration with existing tool registry and configuration system

## Quality Assurance
- Comprehensive test coverage including edge cases
- Domain type definitions for TodoItem and TodoWriteToolResult
- Configuration support with approval settings
- Follows established tool implementation patterns

Closes #42

🤖 Generated with [Claude Code](https://claude.ai/code)

---------

Signed-off-by: Eden Reich <eden.reich@gmail.com>
Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>
Co-authored-by: Eden Reich <edenreich@users.noreply.github.com>

Author: edenreich

.infer/config.yaml

@@ -72,6 +72,9 @@ tools:
       - duckduckgo
       - google
     timeout: 10
+  todo_write:
+    enabled: true
+    require_approval: false
   safety:
     require_approval: true
   exclude_paths:

config/config.go

@@ -44,6 +44,7 @@ type ToolsConfig struct {
 	Tree         TreeToolConfig      `yaml:"tree"`
 	Fetch        FetchToolConfig     `yaml:"fetch"`
 	WebSearch    WebSearchToolConfig `yaml:"web_search"`
+	TodoWrite    TodoWriteToolConfig `yaml:"todo_write"`
 	Safety       SafetyConfig        `yaml:"safety"`
 	ExcludePaths []string            `yaml:"exclude_paths"`
 }
@@ -109,6 +110,12 @@ type WebSearchToolConfig struct {
 	RequireApproval *bool    `yaml:"require_approval,omitempty"`
 }
 
+// TodoWriteToolConfig contains TodoWrite-specific tool settings
+type TodoWriteToolConfig struct {
+	Enabled         bool  `yaml:"enabled"`
+	RequireApproval *bool `yaml:"require_approval,omitempty"`
+}
+
 // ToolWhitelistConfig contains whitelisted commands and patterns
 type ToolWhitelistConfig struct {
 	Commands []string `yaml:"commands"`
@@ -233,6 +240,10 @@ func DefaultConfig() *Config {
 				Engines:       []string{"duckduckgo", "google"},
 				Timeout:       10,
 			},
+			TodoWrite: TodoWriteToolConfig{
+				Enabled:         true,
+				RequireApproval: &[]bool{false}[0],
+			},
 			Safety: SafetyConfig{
 				RequireApproval: true,
 			},
@@ -370,6 +381,10 @@ func (c *Config) IsApprovalRequired(toolName string) bool {
 		if c.Tools.WebSearch.RequireApproval != nil {
 			return *c.Tools.WebSearch.RequireApproval
 		}
+	case "TodoWrite":
+		if c.Tools.TodoWrite.RequireApproval != nil {
+			return *c.Tools.TodoWrite.RequireApproval
+		}
 	}
 
 	return globalApproval

internal/app/chat_application.go

@@ -6,7 +6,7 @@ import (
 	"strings"
 	"time"
 
-	"github.com/charmbracelet/bubbletea"
+	tea "github.com/charmbracelet/bubbletea"
 	"github.com/charmbracelet/lipgloss"
 	"github.com/inference-gateway/cli/internal/container"
 	"github.com/inference-gateway/cli/internal/domain"
@@ -817,7 +817,11 @@ func (app *ChatApplication) handleApprovalKeys(keyMsg tea.KeyMsg) tea.Cmd {
 					}
 				},
 				func() tea.Msg {
-					return handlers.ProcessNextToolCallMsg{}
+					if remainingCalls, ok := app.state.Data["remainingToolCalls"].([]sdk.ChatCompletionMessageToolCall); ok && len(remainingCalls) > 0 {
+						return handlers.ProcessNextToolCallMsg{}
+					} else {
+						return handlers.TriggerFollowUpLLMCallMsg{}
+					}
 				},
 			)()
 		}
@@ -907,7 +911,11 @@ func (app *ChatApplication) approveToolCall() tea.Cmd {
 				}
 			},
 			func() tea.Msg {
-				return handlers.ProcessNextToolCallMsg{}
+				if remainingCalls, ok := app.state.Data["remainingToolCalls"].([]sdk.ChatCompletionMessageToolCall); ok && len(remainingCalls) > 0 {
+					return handlers.ProcessNextToolCallMsg{}
+				} else {
+					return handlers.TriggerFollowUpLLMCallMsg{}
+				}
 			},
 		)()
 	}
@@ -974,7 +982,11 @@ func (app *ChatApplication) denyToolCall() tea.Cmd {
 				}
 			},
 			func() tea.Msg {
-				return handlers.ProcessNextToolCallMsg{}
+				if remainingCalls, ok := app.state.Data["remainingToolCalls"].([]sdk.ChatCompletionMessageToolCall); ok && len(remainingCalls) > 0 {
+					return handlers.ProcessNextToolCallMsg{}
+				} else {
+					return handlers.TriggerFollowUpLLMCallMsg{}
+				}
 			},
 		)()
 	}

internal/domain/interfaces.go

@@ -253,3 +253,19 @@ type DeleteToolResult struct {
 	WildcardExpanded  bool     `json:"wildcard_expanded"`
 	Errors            []string `json:"errors,omitempty"`
 }
+
+// TodoItem represents a single todo item
+type TodoItem struct {
+	ID      string `json:"id"`
+	Content string `json:"content"`
+	Status  string `json:"status"`
+}
+
+// TodoWriteToolResult represents the result of a TodoWrite operation
+type TodoWriteToolResult struct {
+	Todos          []TodoItem `json:"todos"`
+	TotalTasks     int        `json:"total_tasks"`
+	CompletedTasks int        `json:"completed_tasks"`
+	InProgressTask string     `json:"in_progress_task,omitempty"`
+	ValidationOK   bool       `json:"validation_ok"`
+}

internal/handlers/chat_handler.go

@@ -11,7 +11,7 @@ import (
 	"strings"
 	"time"
 
-	"github.com/charmbracelet/bubbletea"
+	tea "github.com/charmbracelet/bubbletea"
 	"github.com/inference-gateway/cli/config"
 	"github.com/inference-gateway/cli/internal/commands"
 	"github.com/inference-gateway/cli/internal/domain"
@@ -64,6 +64,8 @@ func (h *ChatMessageHandler) CanHandle(msg tea.Msg) bool {
 		return true
 	case ProcessNextToolCallMsg:
 		return true
+	case TriggerFollowUpLLMCallMsg:
+		return true
 	case domain.ChatStartEvent, domain.ChatChunkEvent, domain.ChatCompleteEvent, domain.ChatErrorEvent:
 		return true
 	default:
@@ -91,6 +93,9 @@ func (h *ChatMessageHandler) Handle(msg tea.Msg, state *AppState) (tea.Model, te
 	case ProcessNextToolCallMsg:
 		return h.handleProcessNextToolCall(msg, state)
 
+	case TriggerFollowUpLLMCallMsg:
+		return h.handleTriggerFollowUpLLMCall(msg, state)
+
 	case domain.ChatStartEvent:
 		return h.handleChatStart(msg, state)
 
@@ -523,6 +528,9 @@ type StoreRemainingToolCallsMsg struct {
 // ProcessNextToolCallMsg triggers processing of the next tool call in the queue
 type ProcessNextToolCallMsg struct{}
 
+// TriggerFollowUpLLMCallMsg triggers the follow-up LLM call after all tools are executed
+type TriggerFollowUpLLMCallMsg struct{}
+
 // SwitchModelMsg indicates that model selection view should be shown
 type SwitchModelMsg struct{}
 
@@ -820,7 +828,13 @@ func (h *ChatMessageHandler) handleProcessNextToolCall(msg ProcessNextToolCallMs
 	remainingCalls, ok := state.Data["remainingToolCalls"].([]sdk.ChatCompletionMessageToolCall)
 	if !ok || len(remainingCalls) == 0 {
 		delete(state.Data, "remainingToolCalls")
-		return nil, h.triggerFollowUpLLMCall()
+
+		return nil, func() tea.Msg {
+			return shared.SetStatusMsg{
+				Message: "All tool calls completed, preparing follow-up request...",
+				Spinner: true,
+			}
+		}
 	}
 
 	nextCall := remainingCalls[0]
@@ -855,10 +869,13 @@ func (h *ChatMessageHandler) handleProcessNextToolCall(msg ProcessNextToolCallMs
 // triggerFollowUpLLMCall sends the conversation with tool results back to the LLM for reasoning
 func (h *ChatMessageHandler) triggerFollowUpLLMCall() tea.Cmd {
 	return func() tea.Msg {
-		// Convert conversation to SDK messages
 		messages := h.conversationToSDKMessages()
 
-		// Start a new chat completion request
 		return h.startChatCompletion(messages)()
 	}
 }
+
+// handleTriggerFollowUpLLMCall handles the trigger for follow-up LLM call
+func (h *ChatMessageHandler) handleTriggerFollowUpLLMCall(msg TriggerFollowUpLLMCallMsg, state *AppState) (tea.Model, tea.Cmd) {
+	return nil, h.triggerFollowUpLLMCall()
+}

internal/services/tools/registry.go

@@ -32,6 +32,7 @@ func (r *Registry) registerTools() {
 	r.tools["Delete"] = NewDeleteTool(r.config)
 	r.tools["Grep"] = NewGrepTool(r.config)
 	r.tools["Tree"] = NewTreeTool(r.config)
+	r.tools["TodoWrite"] = NewTodoWriteTool(r.config)
 
 	if r.config.Tools.Fetch.Enabled {
 		r.tools["Fetch"] = NewFetchTool(r.config)