OpenAgentsControl — Plan-First AI Agent Framework with Approval Gates

Introduction

OpenAgentsControl adds a structured approval layer to AI coding agents, ensuring that generated code is reviewed, tested, and validated before being applied. It addresses the safety gap in autonomous code generation by requiring explicit plan approval at each step.

What OpenAgentsControl Does

• Generates execution plans before writing any code
• Requires human approval at configurable gates before proceeding
• Runs automatic tests and validation after each code generation step
• Supports multi-language projects across TypeScript, Python, Go, and Rust
• Provides a web dashboard for monitoring agent activity and approving plans

Architecture Overview

The framework wraps AI coding agents in a plan-approve-execute loop. When an agent receives a task, it first generates a structured plan describing the changes. The plan is presented to the user through a web interface. Upon approval, the agent executes the plan and runs automated validation including type checking, linting, and test suites. Failed validations trigger re-planning without manual intervention.

Self-Hosting & Configuration

• Clone the repository and install Node.js dependencies
• Configure agent backends through environment variables
• Set approval policies in the configuration file to control gate strictness
• Deploy the web dashboard locally or on a server for team access
• Supports integration with existing CI pipelines for validation steps

Key Features

• Plan-first workflow preventing unreviewed code changes
• Configurable approval gates from permissive to strict
• Automatic test execution and code review after generation
• Multi-language support with language-specific validation rules
• Web-based dashboard for team collaboration on agent tasks

Comparison with Similar Tools

• Claude Code — powerful coding agent without built-in approval gates; OpenAgentsControl adds structured review
• Aider — AI pair programming focused on conversation; OpenAgentsControl enforces plan-then-execute discipline
• SWE-agent — autonomous issue resolution without approval steps; OpenAgentsControl requires human sign-off
• CrewAI — multi-agent orchestration for general tasks; OpenAgentsControl specializes in code generation safety

FAQ

Q: Does this replace my coding agent? A: No. It wraps existing agents with an approval and validation layer. It works alongside tools you already use.

Q: Can I skip approval for low-risk changes? A: Yes. Approval policies are configurable, allowing auto-approval for changes that pass all automated checks.

Q: What happens if validation fails? A: The agent re-plans automatically, incorporating the validation failure as feedback. After configurable retries, it escalates to human review.

Q: Is this suitable for production use? A: It is designed for development workflows. The approval gates add safety but the framework itself should be evaluated for your team's needs.

Sources

• https://github.com/darrenhinde/OpenAgentsControl