Deterministic Rule based auto-approver for Claude/Codex (github.com)

A new tool called anumati has been introduced as a deterministic auto-approver for AI coding agents, such as Claude Code and OpenAI Codex. This innovative solution efficiently auto-allows safe shell commands using a basic configuration of explicitly defined matchers, streamlining the process by reducing repetitive permission prompts for common commands like `git status` and `cargo test`. Unlike traditional approaches relying on probabilistic assessments from large language models (LLMs), anumati employs strict grammar rules which guarantee consistent decision-making for command approvals. The significance of anumati lies in its ability to enhance security and workflow for developers utilizing AI coding agents. By ensuring that commands are evaluated strictly based on predefined safe categories while still allowing customization through an evolving config, it allows developers to adapt their approval settings based on actual usage. The tool’s architecture supports command line composition, meaning that it only approves entire commands if every component meets safety criteria, thus preventing potentially harmful commands from being executed via chaining. With features like a self-building configuration and integrated suggestions for future command approvals, anumati empowers developers to continuously tailor and secure their command execution environment.