This project provides an autonomous agent orchestrator designed to automate iterative development tasks while users sleep. It addresses the need for continuous progress on objectives by running AI agents in a loop, making small, committed, and documented changes. The primary benefit is waking up to a clean Git branch filled with completed work and a detailed log of the process, making it ideal for developers, researchers, and power users seeking to leverage AI for automated code improvement or task completion.

How It Works

gnhf operates as an autoresearch-style orchestrator. It initiates a loop that validates the Git repository's state, creates a dedicated gnhf/ branch, and iteratively prompts a configured AI agent. Each iteration builds upon context from a notes.md file, executes the agent in non-interactive mode, and either commits successful changes or performs a Git reset with exponential backoff on failure. This approach ensures robust execution, atomic commits, and a clear, traceable history of automated work.

Quick Start & Requirements

• Installation: Install via npm: npm install -g gnhf. Alternatively, build from source following provided instructions.
• Prerequisites: Requires Node.js/npm and Git. Users must also install and configure specific AI agents (e.g., Anthropic's claude CLI, OpenAI's codex CLI, Atlassian's acli for Rovo Dev, or opencode).
• Environment: Runs on macOS, Linux, and Windows within a Git repository with a clean working tree.
• Configuration: Agent settings and paths can be managed via ~/.gnhf/config.yml.
• Links: Detailed CLI reference and flags are available in the README.

Highlighted Details

• Autonomous Iteration: Executes AI agents in a loop, making small, committed changes towards a user-defined objective.
• Robust State Management: Commits successful iterations to a dedicated branch, rolls back failures with exponential backoff, and supports resuming interrupted runs.
• Agent Agnosticism: Integrates with multiple AI agents including Claude Code, Codex, Rovo Dev, and OpenCode via their respective CLIs.
• Configurable Runtime: Offers flags like --max-iterations and --max-tokens to cap execution duration, preventing indefinite runs.
• Detailed Logging: Generates JSONL debug logs capturing lifecycle events, timings, and error details for troubleshooting.

Maintenance & Community

The provided README focuses heavily on technical implementation and lacks explicit details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmaps.

Licensing & Compatibility

The project README does not specify a software license, which prevents an immediate assessment of its compatibility for commercial use or the creation of derivative works.

Limitations & Caveats

The project README does not specify a software license, hindering assessment of commercial use or derivative works. Setup requires significant prerequisite work, including installing and configuring multiple AI agent CLIs and ensuring Git is initialized. The effectiveness and reliability are contingent on the chosen agent's capability to operate consistently in non-interactive modes and adhere strictly to instructions.