Introduction

Beans is an innovative task tracker designed specifically for AI coding agents. Utilizing plain Markdown files, it focuses on verification gates, dependency-aware scheduling, and built-in agent orchestration to streamline project management. Instead of relying on an honor system, every task includes a shell command that must succeed to close the task, ensuring accountability and correctness.

Core Features

• Verification Gates: Every task mandates a shell command to verify its success, emphasizing accountability.
• Fail-First TDD: Tasks must initially fail to guarantee the validity of the test, thus fostering a reliable testing process.
• Dependency-Aware Scheduling: The scheduling is designed to automatically infer dependencies, allowing tasks to unlock as prerequisites are satisfied.
• Agent Orchestration: Built-in functionality allows agents to execute tasks efficiently, with capabilities for monitoring agent performance and managing task execution in parallel.

How It Works

Tasks in Beans are represented as Markdown files with a YAML front matter that defines attributes such as task ID, title, status, and verification command. Here’s a structural overview:

---
id: "1"
title: Add CSV export
status: in_progress
verify: cargo test csv::export
attempts: 0
---

When a task is executed, Beans runs the verify command and tracks attempts, appending failures to the task notes for transparency and accountability. This accumulative context aids later agents in avoiding repeated mistakes.

Usage Example

The following commands illustrate how to leverage Beans for task management:

bn create "Add CSV export" --verify "cargo test csv::export"   # Create a new task with a verification rule
bn run                                                    # Dispatch tasks to agents
bn close 1                                               # Perform a verification to close task ID 1 if successful

With Beans, agents can autonomously read tasks, implement solutions, and report back. This seamless integration fosters a productive environment for continuous delivery and improvement in code quality.

Intelligent Task Management

Beans can manage hierarchical tasks, create smart dependencies, and implement a memory system for ongoing project intelligence. This innovative framework allows teams to accumulate project knowledge stored as verified facts, which can be referenced and verified over time.

Agent Workflow

Beans supports various workflows, including continuous operations and manual task handling. Through features like loop mode and adversarial reviews, it balances efficiency with thorough verification, ensuring a high standard for completed tasks.

Example of Creating and Running Tasks

To initiate the task management process, you can use the following commands:

bn init                           # Initialize the Beans environment
bn create "Fix CSV export" --verify "cargo test csv"  # Create a task with verification
bn run                          # Start dispatching tasks to agents

Conclusion

With Beans, teams can effectively manage workflows while ensuring code quality by relying on measurable, verifiable tasks. This framework offers a significant advantage for developers and organizations looking to integrate AI agents into their task management practices.

For more detailed insights and features, refer to the documentation.

Explore the project and enhance productivity while maintaining rigorous development standards with Beans.