AIAny
AI Agent2026
Icon for item

gnhf

Runs an LLM-driven agent loop that iteratively proposes, applies, tests, and commits small repo changes—each successful iteration becomes a separate git commit while failures are rolled back or preserved for repair. Supports multiple agent backends, worktrees for concurrency, live terminal status, and optional per-iteration pushes.

Introduction

Most agent tools return a result; this one returns a reviewable branch full of incremental, audited changes you can wake up to. The core idea is operationalizing autonomous agents as safe, small, repeatable git iterations so you get traceable progress instead of a single opaque output.

What Sets It Apart
  • Iteration-as-commit: each successful agent iteration becomes a separate unsigned git commit, producing a linear, reviewable history of autonomous changes rather than a single state-diff blob. This makes it straightforward to inspect, revert, or cherry-pick specific agent decisions.
  • Robust failure semantics: non-destructive handling where failed iterations are rolled back, commit-failures are preserved for manual repair, and retryable hard errors back off exponentially while agent-reported failures are retried immediately. Runs end with a permanent exit summary covering elapsed time, iterations, token usage and branch diff stats.
  • Agent- and workflow-agnostic: supports multiple built-in targets (Claude Code, Codex, Rovo Dev, OpenCode, GitHub Copilot CLI, Pi) and ACP targets via a bundled runtime; offers worktree mode to run multiple agents concurrently and flags to push per-iteration or use the current branch.
  • Lightweight operational controls: CLI flags for max-iterations, max-tokens, stop-when conditions, prevent-sleep, live terminal title updates, and detailed local run metadata under .gnhf for auditing and resume.
Who it's for + tradeoffs

Great fit if you want continuous, reviewable automation: maintainers who accept lots of small, well-documented changes (CI-driven projects, experiment loops, refactoring by automated agents) and value audit trails for each autonomous action. It excels at long-running unattended runs where incremental progress and recoverability matter.

Look elsewhere if you need a single large atomic refactor, handle large binary artifacts, or require GUI-driven orchestration—gnhf is opinionated about small, testable commits and requires a clean git repository and CI/tests to validate agent changes. It also assumes comfort with CLI workflows and Node/TypeScript tooling for installation and extension.

Information

  • Websitegithub.com
  • Authorskunchenguid
  • Published date2026/03/31

Categories

More Items

Hugging Face

Provides 5.5K+ self-contained data-analysis RL tasks: each row bundles a real tabular dataset, a question, and a deterministically-gradable gold answer. Verified from jupyter-agent notebooks; splits for training, held-out testing, and quick eval; intended for prompting, fine-tuning, and agent RL.

Hugging Face
AI Model2026

A 9B agentic multimodal SFT checkpoint distilled from Qwen3.5-9B for coding, general agent tasks, visual coding and cybersecurity. Provided by Xiaomi MiMo as a research seed (77.4B-token SFT mix) to bootstrap agentic RL and tool-use experiments.

Hugging Face
AI Model2026

Preview agentic language model for research and engineering workflows that turns research questions into executable, verifiable workflows via tool use and long-context reasoning; built on a 744B-parameter MoE (GLM-5.2) with MIT-licensed BF16 and FP8 checkpoints.