kunchenguid/gnhf

Before I go to bed, I tell my agents: good night, have fun

What it solves

gnhf is an autonomous orchestrator for coding agents that allows them to work on a codebase over long periods (such as overnight) without constant human supervision. It solves the problem of "empty-handed" wake-ups by ensuring agents make small, committed, and documented incremental changes toward a goal, rather than attempting one giant, risky leap.

How it works

The tool wraps existing coding agent CLIs in a loop. For each iteration, it:

  1. Prompts the Agent: Injects a goal and a notes.md file (which acts as shared memory across iterations).
  2. Executes: Runs the agent in non-interactive mode.
  3. Validates & Commits: If the agent succeeds, gnhf creates a git commit and appends to the notes. If it fails, it rolls back the changes using git reset --hard (unless it was a commit failure, which it preserves for repair).
  4. Handles Limits: It automatically manages rate limits (e.g., Claude's usage windows) by waiting for the reset time before resuming.
  5. Manages State: It can run on a dedicated gnhf/ branch, the current branch, or in isolated git worktrees to allow multiple agents to work on the same repo simultaneously.

Who it’s for

Developers who use AI coding agents and want to automate long-running refactors, feature implementations, or codebase cleanups without needing to monitor the terminal every few minutes.

Highlights

  • Agent-Agnostic: Works with popular coding agent CLIs and any ACP target.
  • Incremental Progress: Every successful step is committed to git, allowing for easy cherry-picking or reverts.
  • Worktree Support: Run multiple agents concurrently on the same repository in isolated directories.
  • Smart Rate-Limit Handling: Automatically sleeps through API usage windows and resumes work.
  • Resume Capability: Can pick up a previous run from an existing branch if interrupted.

Related

  • Project
    viettran-edgeAI/codex_workflow

    codex_workflow is a Python‑based add‑on for the Codex AI assistant that introduces three execution modes (light, medium, heavy). It automatically creates a persistent `agent_docs/` folder to store project goals, decisions and progress, and it can spawn specialised sub‑agents (Explorer, Investigator, Executor, Tester, Archivist) to share work and reduce the main agent’s token usage. Installation is a manual zip download with a bootstrap script; usage is as simple as typing `use medium route. <task>` in the Codex chat. The tool includes token‑usage reporting and benchmarks showing how its batching heuristics cut costs, but it is limited by Codex’s four‑agent‑per‑session cap and requires manual updates.

  • Project
    Intelligent-Internet/zenith

    Zenith is a Python harness that lets Claude Code, Codex, or Hermes run as long‑running coding agents. An orchestrator repeatedly checks project state, spawns worker/tester sub‑agents, registers reusable skills, replans, and only stops when the task is truly complete. Benchmarked on the Frontier SWE suite, Zenith (using GPT‑5.5) achieves the best average rank while costing less than half of the strongest baseline (RALPH). The repo provides a CLI, MCP server, ACP adapters, and a full technical report.

  • Project
    maddada/Ghostex

    Ghostex is a cross-platform workspace for managing multiple AI coding agents, providing a unified interface for chat, terminal access, and IDE integration that persists across sessions and devices.

  • Dispatch
    Hugging Face hf CLI agent‑optimized redesign cuts token usage and boosts success rates

    Hugging Face announced an agent‑optimized redesign of the hf CLI that cuts token usage by up to 6× and improves success rates for coding agents like Claude Code and Codex.

  • Project
    HangYu8123/HarnessFlow

    HarnessFlow is a portable set of Markdown instruction files that turn Claude Code, Codex CLI, or GitHub Copilot into a multi‑step, self‑reviewing coding assistant. It provides request templates, parallel analysis agents, a devil’s‑advocate challenge, QA validation, and persistent repo memory. Install by copying the pack into `.github/HarnessFlow/` of your repo and running the provided `setup.sh` or `cli_setup.sh`. Choose a request template (code, refactor, debug, etc.), fill it in, and paste it to the assistant. The pack supports three workflow modes—`general`, `fast`, and `skill`—and reports benchmark gains (up to 76 % fewer generated lines while staying 100 % correct).