amplifier-module-tool-todo

AI self-accountability tool for managing todo lists during complex multi-step tasks.

Purpose

Provides cognitive scaffolding for AI agents to maintain focus through long, complex turns where context fills up and plans might get buried. Works with amplifier-module-hooks-todo-reminder for automatic context injection.

This is NOT for showing task progress to users - it's for AI self-management to prevent wandering or forgetting steps.

The Problem It Solves

Typical failure pattern:

1. AI gets complex task, creates plan [Step 1, Step 2, Step 3, Step 4, Step 5]
2. Completes steps 1-2, context fills with results (30k+ tokens)
3. At step 3: AI forgets original plan buried in context
4. Risk: AI wanders, declares done prematurely, abandons steps 4-5

Solution:

• AI creates todo list using this tool
• Hook injects reminders before each turn
• AI sees: "✓1 done, ✓2 done, ☐3 pending, ☐4 pending, ☐5 pending"
• AI stays on track, completes all commitments

Installation

uv add amplifier-module-tool-todo

Configuration

Add to your profile:

tools:
  - module: tool-todo
    source: git+https://github.com/microsoft/amplifier-module-tool-todo@main
    config: {} # No configuration needed

AI Usage Patterns

Pattern 1: Create Plan at Task Start

User: "Implement a new authentication system"

AI thought process:
1. This is complex, I should create a plan
2. Use todo tool to create list
3. Work through systematically

AI action:

{
  "action": "create",
  "todos": [
    {
      "content": "Research auth requirements",
      "activeForm": "Researching auth requirements",
      "status": "pending"
    },
    {
      "content": "Design auth flow",
      "activeForm": "Designing auth flow",
      "status": "pending"
    },
    {
      "content": "Implement auth module",
      "activeForm": "Implementing auth module",
      "status": "pending"
    },
    {
      "content": "Write auth tests",
      "activeForm": "Writing auth tests",
      "status": "pending"
    },
    {
      "content": "Document auth usage",
      "activeForm": "Documenting auth usage",
      "status": "pending"
    }
  ]
}

Pattern 2: Update After Each Step

[AI completes research]

AI action:
```json
{
  "action": "update",
  "todos": [
    {"content": "Research auth requirements", "activeForm": "Researching auth requirements", "status": "completed"},
    {"content": "Design auth flow", "activeForm": "Designing auth flow", "status": "in_progress"},
    {"content": "Implement auth module", "activeForm": "Implementing auth module", "status": "pending"},
    {"content": "Write auth tests", "activeForm": "Writing auth tests", "status": "pending"},
    {"content": "Document auth usage", "activeForm": "Documenting auth usage", "status": "pending"}
  ]
}

Result: AI updates plan, hook will inject reminder in next turn

Pattern 3: Check Status Anytime

{
  "action": "list"
}

Returns: Current todo list with all items and their statuses

Todo Item Schema

Each todo must have:

{
  content: string; // Imperative: "Run tests", "Build project"
  activeForm: string; // Present continuous: "Running tests", "Building project"
  status: "pending" | "in_progress" | "completed";
}

Rules:

• Keep ONE item as "in_progress" at a time (guidance, not enforced)
• Mark "completed" immediately after finishing each step
• Use imperative for content, present continuous for activeForm

Tool Interface

Actions

create: Replace entire list with new todos

{
  "action": "create",
  "todos": [...]
}

update: Replace entire list (AI manages transitions)

{
  "action": "update",
  "todos": [...]  // Full list with updated statuses
}

list: Return current todos

{
  "action": "list"
}

Responses

create/update:

{
  "status": "created|updated",
  "count": 5,
  "pending": 3,
  "in_progress": 1,
  "completed": 1,
  "todos": [...]  // Full list returned
}

list:

{
  "status": "listed",
  "count": 5,
  "todos": [...]
}

How It Works

1. AI calls tool to create/update todo list
2. Tool stores in coordinator.todo_state (session-scoped)
3. Hook watches for prompt:submit events
4. Hook injects formatted reminder into AI's context
5. AI sees reminder before each response
6. AI stays focused on completing all steps

Philosophy Alignment

• ✅ Mechanism not policy: Tool provides storage, hook provides injection, AI decides usage
• ✅ Ruthless simplicity: In-memory, basic CRUD, no complex validation
• ✅ Session-scoped: State lives only during conversation
• ✅ AI self-management: No enforcement, just scaffolding
• ✅ Non-interference: Failures don't crash session

Implementation Details

• Storage: coordinator.todo_state = [] (dynamic attribute)
• Scope: Session-only (cleared when session ends)
• Validation: Schema validation only (required fields, valid statuses)
• Events: Emits TOOL_PRE and TOOL_POST for observability

Testing

Run integration tests:

# Unit test: Tool operations
python test_todo_integration.py

# Multi-turn test: Accountability loop
python test_todo_multi_turn.py

Both tests verify:

• Tool CRUD operations work correctly
• Hook injects reminders on prompt:submit
• Context receives injected messages
• Format matches TodoWrite display (✓/☐ symbols)

Contributing

Note

This project is not currently accepting external contributions, but we're actively working toward opening this up. We value community input and look forward to collaborating in the future. For now, feel free to fork and experiment!

Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit Contributor License Agreements.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.