mt2007

Autopilot Harness — Development skill for Claude Code

Development community

Vibecoding harness: structured planning → checklist execution → multi-lens self-review.

How to install Autopilot Harness

This entry records only its repository, not the path inside it, so there is no exact command to give. Open mt2007/autopilot-harness and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Autopilot Harness does

Vibecoding harness: structured planning → checklist execution → multi-lens self-review. Works with Cursor, Claude Code, Codex, and more.

Alternatives in Development

Understand Anything — Graphs that teach graphs that impress 80.7k ★
Mempalace Init — Set up MemPalace — install the package, initialize a palace, register the MCP server with Cursor, and verify e 58.6k ★
Lazycodex — The one and only agent harness for complex codebases 3.3k ★

README

Autopilot

**autopilot-harness** — a vibecoding harness that turns open-ended agent chat into **structured planning → checklist execution → multi-lens self-review**. ([中文说明](./README.zh-CN.md))

[![CI](https://github.com/mt2007/autopilot-harness/actions/workflows/ci.yml/badge.svg)](https://github.com/mt2007/autopilot-harness/actions/workflows/ci.yml) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](./LICENSE) [![Node.js](https://img.shields.io/badge/Node.js-%E2%89%A522-brightgreen.svg)](https://nodejs.org/)

Autopilot Harness — State the goal. Autopilot the rest.

Why

Vibe coding is fast until scope drifts, acceptance stays implicit, and “looks done” skips hard review angles. Autopilot keeps the agent on a durable track:

Grill the plan before writing product code
Execute against a checklist (artifacts.plans_dir//, new-init default docs/autopilot/plans)
Self-review under rotating lenses before an item is marked done

It is **not** a general-purpose chat agent, **not** a substitute for your CI/test framework, **not** a Jira/kanban product (checklist + execution FSM, not a board UI). **This build ships Cursor, Claude Code, Codex, Kimi Code, GitHub Copilot CLI, Grok Build CLI, Gemini CLI, Factory Droid, Hermes Agent, Antigravity, Pi, Devin CLI, and Runner (meta)** (CLI + App share one Codex port; **Kimi Code** is **degraded Stop≤1/turn**; **Copilot CLI** is **degraded Stop consecutive ≤8**; **Grok Build** is **degraded Stop ≤8/turn** — per-turn reset; **Gemini CLI** is **Shipped** with honest **AfterAgent turn cap ≤100** / `MAX_TURNS` and soft min-CLI **≥0.31.0**; **Factory Droid** is **Shipped** with **multi-block under `stop_hook_active` live-proved** — no raise; **waive live → degraded≤1**; **Hermes Agent** is **Shipped** with **shell `pre_verify` continue live-proved** — nudge ≥32; edit-only; soft min **≥