🤖 Add TODO list feature for tracking multi-step tasks (#241)

Adds TodoWrite and TodoRead tools to enable the AI to manage task lists
during complex operations, following Claude Code's pattern with a
simplified schema.

## Changes

**Tool Implementation:**
- TodoWrite: Full-replacement semantics (replaces entire list on each
call)
- TodoRead: Returns current todo list
- Stream-scoped filesystem storage in tmpdir (automatic cleanup when
stream ends)
- Schema: content, status (pending/in_progress/completed), activeForm
fields

**UI Component:**
- Visual indicators: ✓ (completed), ⏳ (in_progress), ○ (pending)
- Shows activeForm text for in-progress items
- Color-coded borders and backgrounds by status

**Testing:**
- Unit tests for storage layer (7 tests)
- All existing tests pass (477 total)

## Design Decisions

**Simplified schema:** Removed id, priority, timestamps, and metadata
fields (unnecessary complexity). Added `activeForm` for better UX
("Adding tests" vs "Add tests").

**Stream-scoped storage:** TODOs stored in stream's tmpdir as
`todos.json`. Automatically cleaned up when stream ends. Each stream has
isolated storage (no cross-stream pollution).

**Full-replacement semantics:** AI replaces entire list on each call.
Simpler than incremental updates and matches Claude Code's pattern.

## Benefits

✅ **Automatic cleanup:** tmpdir removed when stream ends, preventing
memory leaks
✅ **Stream isolation:** Concurrent streams in same workspace don't share
TODOs
✅ **Inspectable:** Can view todos.json during debugging  
✅ **Crash resilient:** TODOs persisted to disk during stream  

_Generated with `cmux`_

Author: ammar-agent

src/components/Messages/ToolMessage.tsx

@@ -6,6 +6,7 @@ import { BashToolCall } from "../tools/BashToolCall";
 import { FileEditToolCall } from "../tools/FileEditToolCall";
 import { FileReadToolCall } from "../tools/FileReadToolCall";
 import { ProposePlanToolCall } from "../tools/ProposePlanToolCall";
+import { TodoToolCall } from "../tools/TodoToolCall";
 import type {
   BashToolArgs,
   BashToolResult,
@@ -19,6 +20,8 @@ import type {
   FileEditReplaceLinesToolResult,
   ProposePlanToolArgs,
   ProposePlanToolResult,
+  TodoWriteToolArgs,
+  TodoWriteToolResult,
 } from "@/types/tools";
 
 interface ToolMessageProps {
@@ -65,6 +68,11 @@ function isProposePlanTool(toolName: string, args: unknown): args is ProposePlan
   return TOOL_DEFINITIONS.propose_plan.schema.safeParse(args).success;
 }
 
+function isTodoWriteTool(toolName: string, args: unknown): args is TodoWriteToolArgs {
+  if (toolName !== "todo_write") return false;
+  return TOOL_DEFINITIONS.todo_write.schema.safeParse(args).success;
+}
+
 export const ToolMessage: React.FC<ToolMessageProps> = ({ message, className, workspaceId }) => {
   // Route to specialized components based on tool name
   if (isBashTool(message.toolName, message.args)) {
@@ -144,6 +152,18 @@ export const ToolMessage: React.FC<ToolMessageProps> = ({ message, className, wo
     );
   }
 
+  if (isTodoWriteTool(message.toolName, message.args)) {
+    return (
+      <div className={className}>
+        <TodoToolCall
+          args={message.args}
+          result={message.result as TodoWriteToolResult | undefined}
+          status={message.status}
+        />
+      </div>
+    );
+  }
+
   // Fallback to generic tool call
   return (
     <div className={className}>

src/components/tools/TodoToolCall.tsx

@@ -0,0 +1,161 @@
+import React from "react";
+import styled from "@emotion/styled";
+import type { TodoWriteToolArgs, TodoWriteToolResult, TodoItem } from "@/types/tools";
+import {
+  ToolContainer,
+  ToolHeader,
+  ExpandIcon,
+  StatusIndicator,
+  ToolDetails,
+} from "./shared/ToolPrimitives";
+import { useToolExpansion, getStatusDisplay, type ToolStatus } from "./shared/toolUtils";
+import { TooltipWrapper, Tooltip } from "../Tooltip";
+
+const TodoList = styled.div`
+  display: flex;
+  flex-direction: column;
+  gap: 3px;
+  padding: 6px 8px;
+`;
+
+const TodoItemContainer = styled.div<{ status: TodoItem["status"] }>`
+  display: flex;
+  align-items: flex-start;
+  gap: 6px;
+  padding: 4px 8px;
+  background: ${(props) => {
+    switch (props.status) {
+      case "completed":
+        return "color-mix(in srgb, #4caf50, transparent 92%)";
+      case "in_progress":
+        return "color-mix(in srgb, #2196f3, transparent 92%)";
+      case "pending":
+      default:
+        return "color-mix(in srgb, #888, transparent 96%)";
+    }
+  }};
+  border-left: 2px solid
+    ${(props) => {
+      switch (props.status) {
+        case "completed":
+          return "#4caf50";
+        case "in_progress":
+          return "#2196f3";
+        case "pending":
+        default:
+          return "#666";
+      }
+    }};
+  border-radius: 3px;
+  font-family: var(--font-monospace);
+  font-size: 11px;
+  line-height: 1.35;
+  color: var(--color-text);
+`;
+
+const TodoIcon = styled.div`
+  font-size: 12px;
+  flex-shrink: 0;
+  margin-top: 1px;
+  opacity: 0.8;
+`;
+
+const TodoContent = styled.div`
+  flex: 1;
+  min-width: 0;
+`;
+
+const TodoText = styled.div<{ status: TodoItem["status"] }>`
+  color: ${(props) => (props.status === "completed" ? "#888" : "var(--color-text)")};
+  text-decoration: ${(props) => (props.status === "completed" ? "line-through" : "none")};
+  opacity: ${(props) => (props.status === "completed" ? "0.7" : "1")};
+`;
+
+const TodoActiveForm = styled.div`
+  color: #2196f3;
+  font-weight: 500;
+  font-size: 11px;
+  opacity: 0.95;
+  white-space: nowrap;
+
+  &::after {
+    content: "...";
+    display: inline;
+    overflow: hidden;
+    animation: ellipsis 1.5s steps(4, end) infinite;
+  }
+
+  @keyframes ellipsis {
+    0% {
+      content: "";
+    }
+    25% {
+      content: ".";
+    }
+    50% {
+      content: "..";
+    }
+    75% {
+      content: "...";
+    }
+  }
+`;
+
+interface TodoToolCallProps {
+  args: TodoWriteToolArgs;
+  result?: TodoWriteToolResult;
+  status?: ToolStatus;
+}
+
+function getStatusIcon(status: TodoItem["status"]): string {
+  switch (status) {
+    case "completed":
+      return "✓";
+    case "in_progress":
+      return "⏳";
+    case "pending":
+    default:
+      return "○";
+  }
+}
+
+export const TodoToolCall: React.FC<TodoToolCallProps> = ({
+  args,
+  result: _result,
+  status = "pending",
+}) => {
+  const { expanded, toggleExpanded } = useToolExpansion(true); // Expand by default
+  const statusDisplay = getStatusDisplay(status);
+
+  return (
+    <ToolContainer expanded={expanded}>
+      <ToolHeader onClick={toggleExpanded}>
+        <ExpandIcon expanded={expanded}>▶</ExpandIcon>
+        <TooltipWrapper inline>
+          <span>📋</span>
+          <Tooltip>todo_write</Tooltip>
+        </TooltipWrapper>
+        <StatusIndicator status={status}>{statusDisplay}</StatusIndicator>
+      </ToolHeader>
+
+      {expanded && (
+        <ToolDetails>
+          <TodoList>
+            {args.todos.map((todo, index) => (
+              <TodoItemContainer key={index} status={todo.status}>
+                <TodoIcon>{getStatusIcon(todo.status)}</TodoIcon>
+                <TodoContent>
+                  {todo.status === "in_progress" ? (
+                    <TodoActiveForm>{todo.activeForm}</TodoActiveForm>
+                  ) : (
+                    <TodoText status={todo.status}>{todo.content}</TodoText>
+                  )}
+                </TodoContent>
+              </TodoItemContainer>
+            ))}
+          </TodoList>
+        </ToolDetails>
+      )}
+    </ToolContainer>
+  );
+};