---
id: manual-valkor-ai-loom
name: "valkor-ai/loom"
nameZh: "valkor-ai/loom"
sourceCategory: "Loop Engineering Skills"
sourceCategoryZh: "Loop Engineering Skills"
url: "https://github.com/valkor-ai/loom"
sourceUrl: "https://github.com/search?q=loops+engineering&type=repositories"
tags: ["Loop engineering","Coding agents","Delivery harness","Codex"]
tagsZh: ["Loop engineering","Coding agents","Delivery harness","Codex"]
categorySlugs: ["dev","testing","meta-tools","productivity"]
---

# valkor-ai/loom

中文名称：valkor-ai/loom

## Description

Loom is a loop engineering delivery harness for turning Codex, Claude Code, OpenCode, and other coding agents into repeatable software delivery systems.

Loom 是一个 loop engineering 交付 harness，用于把 Codex、Claude Code、OpenCode 等 coding agent 组织成可重复的软件交付系统。

## Why It Was Selected

This repository was selected from the GitHub search results for `loops engineering` because it is not just an essay or concept page. It provides a project-local delivery layer, adapter commands, state persistence, task routing, verification, repair, preview, and handoff flow for coding agents.

GitHub API snapshot: non-fork repository, TypeScript, 286 stars, updated 2026-06-16 UTC.

## Use This Skill

- Open or install the skill from: https://github.com/valkor-ai/loom
- Review the GitHub search source: https://github.com/search?q=loops+engineering&type=repositories
- Categories: dev, testing, meta-tools, productivity
- Tags: Loop engineering, Coding agents, Delivery harness, Codex
- 中文标签：Loop engineering、Coding agents、Delivery harness、Codex

## README SEO Report

### Detailed Overview

Loom describes itself as an open-source delivery harness for existing coding agents. It turns a delivery goal into a structured loop of planning, building, verification, repair, preview, and handoff. The repository is especially relevant for loop engineering because it persists project context, task contracts, backend state, test results, preview evidence, repair notes, and handoff reports under `.loom/`.

The key workflow is not a one-shot prompt. Loom routes the next step, executes work, verifies the result, records evidence, repairs when needed, and continues from saved state. That makes it a useful reference for teams designing agent delivery loops that need to survive interruption, compaction, adapter switches, and future handoffs.

### Installation

The README lists Node.js 20+, npm, a supported coding agent CLI, and Docker for deployment previews. It provides adapter install scripts for Codex, Claude Code, OpenCode, and all adapters:

```bash
npm run plugin:install-codex
npm run plugin:install-claude
npm run plugin:install-opencode
npm run plugin:install-adapters
```

### Usage

Use Loom when a request is larger than a one-shot edit and needs clarification, architecture, task splitting, implementation evidence, review, repair, preview, deployment, or handoff.

Example agent commands from the README:

```text
@loom build a visitor registration system
@loom plan this feature first
@loom continue
@loom review
@loom deploy
```

### Capabilities

- Dynamic workflow routing for coding-agent delivery.
- Project-local `.loom/` state.
- Task contracts and evidence records.
- Verification, repair, preview, and handoff steps.
- Codex, Claude Code, and OpenCode adapter surfaces.

### Source Archive

- Archive source type: generated-local-file
- Archived docs: None

## 3W FAQ

### What is this skill?

Loom is a loop engineering delivery harness for software delivery with coding agents.

### Who is this skill for?

This skill is for developers and AI agent teams working on development, testing, meta-tools, and productivity workflows.

### When should you use it?

Use it when a coding-agent task needs durable state, verification, repair, preview evidence, or handoff instead of a one-shot prompt.

## 中文 3W FAQ

### What：这个 skill 是什么？

Loom 是面向 coding agent 的 loop engineering 交付 harness，把需求澄清、计划、实现、验证、修复、预览和交接变成可重复流程。

### Who：适合谁使用？

适合处理 development、testing、meta-tools、productivity 工作流的开发者、团队和 AI Agent。

### When：什么时候使用？

当任务需要跨多轮保留上下文、运行验证、记录证据、修复失败并交付可检查结果时使用。