Chappygo-OS

Atomic Spec — Testing skill for Claude Code

Testing community

Stop your AI from vibe-coding.

How to install Atomic Spec

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

What Atomic Spec does

Stop your AI from vibe-coding. Governance layer that forces Claude, Cursor, Copilot, Gemini & 13 more through gated phases, atomic task files, and pinned context.

Alternatives in Testing

AI Contribution Skills (Engram) — branch-pr: clean branch and PR workflow 2.5k ★
Spectra App — Spec-driven development for coding agents — a desktop app, a CLI, and skills for Codex, Claude Code, Cursor, C 735 ★
Spec Kit Zh — 💫 帮助你快速上手规格驱动开发的工具包💫 Toolkit to help you get started with Spec-Driven Development Works seamlessly across al 238 ★

README

Stop your AI from vibe-coding.

[![PyPI](https://img.shields.io/pypi/v/atomic-spec.svg?color=blue&logo=pypi&logoColor=white)](https://pypi.org/project/atomic-spec/) [![Python](https://img.shields.io/pypi/pyversions/atomic-spec.svg?logo=python&logoColor=white)](https://pypi.org/project/atomic-spec/) [![License](https://img.shields.io/badge/license-MIT-blue.svg)](./LICENSE) [![Agents supported](https://img.shields.io/badge/agents-17-8A2BE2)](#ai-coding-agents-supported) [![Status](https://img.shields.io/badge/status-v0.5.x%20%E2%80%94%20APIs%20may%20change-yellow)](#status)

**Spec-kit taught AI agents a workflow. Atomic Spec makes them obey it.** When `/atomicspec.implement` runs, the agent is *architecturally prevented* from reading `plan.md` or `spec.md` — it sees only the current task file, pre-loaded with every registry value, domain rule, and gate criterion it needs. No drifting mid-feature. No hallucinating a new approach on pass three. No 800-line "kitchen sink" PRs.

uv tool install atomic-spec
atomicspec init my-project --ai claude

**TL;DR** — One AI instruction file isn't enough. Atomic Spec enforces **nine non-negotiable rules**: one task per file, AI sees only the current task during coding, human sign-off at four checkpoints, a project registry that remembers every decision, and cross-provider handoff detection so a fresh session picks up exactly where the last one crashed. If you've ever reviewed an AI PR and thought *"where did this pattern come from?"* — that's the problem this exists to kill.

Status

**v0.5.0 · current.** The core governance model (Nine Prime Directives — Directive 9 added in v0.3 for the Orientation Read Surface) is stable and exercised on every release. APIs and command names may evolve through v0.x. Claude Code is the reference implementation and validated end-to-end; 12 other agents are experimental (