experimental: add todo list tool

[ElasticBottle]

No description provided.

[pkg-pr-new]

Open in StackBlitz

@trymeka/ai-provider-vercel

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/ai-provider-vercel@53

@trymeka/computer-provider-anchor-browser

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/computer-provider-anchor-browser@53

@trymeka/computer-provider-core

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/computer-provider-core@53

@trymeka/computer-provider-e2b

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/computer-provider-e2b@53

@trymeka/computer-provider-scrapybara

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/computer-provider-scrapybara@53

@trymeka/core

pnpm add https://pkg.pr.new/trymeka/agent/@trymeka/core@53

commit: 763220f

[greptile-apps]

Greptile Summary

This PR introduces an experimental todo list tool to the Meka Agent system, enabling AI agents to break down complex tasks into manageable subtasks with progress tracking. The implementation follows the existing tool architecture pattern used by the memory and computer tools.

The core changes include:

1. New Todo List Tool: A complete implementation in packages/core/src/tools/todo-list.ts with Zod schema validation supporting three actions (add, update, list) and task status tracking (pending, in-progress, completed, cancelled)

2. Session-based Storage: The SessionToDoListStore class provides in-memory task persistence within a single agent session using a Map-based approach with auto-generated task IDs

3. Agent Integration: The todo list tool is integrated into the core agent system (packages/core/src/ai/agent.ts) alongside existing tools, with automatic context injection that prepends current task list state to conversation messages

4. System Prompt Documentation: Updated the AI system prompt to inform the agent about the todo list tool's capabilities and when to use it for complex task planning

5. Example Modification: The anchor-browser example was updated to test the new functionality by removing structured output requirements and changing the task to analyze commonalities among Hacker News posts

The tool integrates seamlessly with the existing architecture, using the same session-scoped pattern as the memory tool and following established conventions for tool creation, logging, and context injection. This enhancement addresses a key limitation where agents could lose track of progress in multi-step workflows.

Confidence score: 4/5

• This PR appears safe to merge with the experimental todo list functionality working as intended
• The implementation follows established patterns and integrates well with existing architecture, though it's marked experimental
• The anchor-browser example changes need attention as they comment out functionality rather than properly removing it

Context used:

Rule - Try to find real bugs instead of nit picking on small style issues. We are moving fast and trying to make sure things work. (link)

_{5 files reviewed, no comments}

_{Edit Code Review Bot Settings | Greptile}