fix(todo): preserve compatibility outside active plans

packages/core/src/core/prompts.test.ts

@@ -168,6 +168,12 @@ describe('Core System Prompt (prompts.ts)', () => {
     expect(prompt).toContain(
       'rather than one item per error, file, command, or minor edit',
     );
+    expect(prompt).toContain(
+      'When an active Todo plan covers work delegated through top-level Agent calls',
+    );
+    expect(prompt).not.toContain(
+      'For complex work delegated through top-level Agent calls, create the relevant todo first',
+    );
     expect(prompt).not.toContain('VERY frequently');
     expect(prompt).not.toContain('EXTREMELY helpful');
     expect(prompt).not.toContain('write 10 items to the todo list');

packages/core/src/core/prompts.ts

@@ -317,7 +317,7 @@ You have access to the ${ToolNames.TODO_WRITE} tool to keep user-visible progres
 
 When you create a todo list:
 - Keep it short and outcome-oriented. Use a few meaningful, logically ordered, verifiable steps rather than one item per error, file, command, or minor edit.
-- For complex work delegated through top-level Agent calls, create the relevant todo first and pass its ID as \`todo_id\` so the execution can be associated with the plan node.
+- When an active Todo plan covers work delegated through top-level Agent calls, pass the matching Todo ID as \`todo_id\` so the execution can be associated with that plan node. Do not create a Todo solely to wrap a delegation that does not otherwise need task tracking.
 - Keep at most one item in_progress. Keep the list current, mark finished work completed, and revise it when the scope or approach changes. When work completes together, update multiple statuses in one tool call rather than making bookkeeping-only calls.
 - Do not repeat the full todo list in prose after calling the tool; briefly communicate only important context or the next step.
 

packages/core/src/tools/todoWrite.test.ts

@@ -327,7 +327,7 @@ describe('TodoWriteTool', () => {
 
       const result = await tool
         .build({
-          todos: [{ id: '1', content: 'Still done', status: 'completed' }],
+          todos: [{ id: '1', content: 'Done', status: 'completed' }],
         })
         .execute(mockAbortSignal);
 
@@ -348,7 +348,28 @@ describe('TodoWriteTool', () => {
       mockAtomicWrite.mockResolvedValue(undefined);
 
       const result = await tool
-        .build({ todos: [{ id: '2', content: 'New', status: 'pending' }] })
+        .build({ todos: [{ id: '1', content: 'New', status: 'pending' }] })
+        .execute(mockAbortSignal);
+      const display = result.returnDisplay as { planId?: string };
+
+      expect(display.planId).toEqual(expect.any(String));
+      expect(display.planId).not.toBe('finished-plan');
+    });
+
+    it('should start a new plan for a distinct all-completed snapshot', async () => {
+      mockFs.readFile.mockResolvedValue(
+        JSON.stringify({
+          planId: 'finished-plan',
+          todos: [{ id: '1', content: 'Done', status: 'completed' }],
+        }),
+      );
+      mockFs.mkdir.mockResolvedValue(undefined);
+      mockAtomicWrite.mockResolvedValue(undefined);
+
+      const result = await tool
+        .build({
+          todos: [{ id: '1', content: 'Already done', status: 'completed' }],
+        })
         .execute(mockAbortSignal);
       const display = result.returnDisplay as { planId?: string };
 

packages/core/src/tools/todoWrite.ts

@@ -8,6 +8,7 @@ import type { ToolResult } from './tools.js';
 import { BaseDeclarativeTool, BaseToolInvocation, Kind } from './tools.js';
 import type { FunctionDeclaration } from '@google/genai';
 import { randomUUID } from 'node:crypto';
+import { isDeepStrictEqual } from 'node:util';
 import * as fs from 'fs/promises';
 import * as fsSync from 'fs';
 import * as path from 'path';
@@ -370,7 +371,7 @@ class TodoWriteToolInvocation extends BaseToolInvocation<
         finalTodos.length > 0 &&
         (oldTodos.length === 0 ||
           (oldTodos.every((todo) => todo.status === 'completed') &&
-            finalTodos.some((todo) => todo.status !== 'completed')));
+            !isDeepStrictEqual(finalTodos, oldTodos)));
       const activePlanId =
         finalTodos.length === 0
           ? undefined

packages/web-shell/client/components/messages/PlanExecutionView.test.tsx

@@ -135,6 +135,20 @@ describe('PlanExecutionView', () => {
     });
   });
 
+  it('does not block a todo on an unknown dependency', () => {
+    const todo: TodoItem = {
+      id: 'standalone',
+      content: 'Standalone',
+      status: 'pending',
+      blockedBy: ['missing'],
+    };
+
+    expect(getPlanNodeState(todo, new Map([[todo.id, todo]]), [], [])).toEqual({
+      status: 'ready',
+      attention: false,
+    });
+  });
+
   it('restores cancellation from replay output after the live task leaves', () => {
     const cancelled = {
       ...agentTool('build'),

packages/web-shell/client/components/messages/PlanExecutionView.tsx

@@ -228,7 +228,7 @@ function getPlanNodeStateFromIndex(
     return { status: 'paused', attention };
   if (todo.status === 'completed') return { status: 'completed', attention };
   const blocked = (todo.blockedBy ?? []).some(
-    (id) => todosById.get(id)?.status !== 'completed',
+    (id) => todosById.has(id) && todosById.get(id)?.status !== 'completed',
   );
   if (blocked) return { status: 'blocked', attention };
   if (todo.status === 'in_progress')