什么是 Wreckit Skill (jmanhype 分支)？

Wreckit Skill (jmanhype 分支) 是一项复杂的集成，它将 Clawdbot 连接到专为大规模开发设计的自主软件工厂。此版本专门针对 Cattle 架构进行了优化，利用瞬时 Sprite VM 确保每项任务都在干净、隔离的环境中执行。通过利用此类 Openclaw Skills，开发人员可以在复杂的自主编码循环中实现最高的安全性并防止状态污染。

该技能为管理整个软件开发生命周期提供了一个强大的框架，从最初的构思和路线图制定到自愈诊断和模式提取。无论是在使用 Firecracker 微型虚拟机（microVM）的安全 Cattle 模式下运行，还是在高速本地 Pet 模式下运行，Wreckit 都能赋予 AI 智能体作为全功能软件工程师的能力，并具备先进的回滚和优化功能。

下载入口:https://github.com/openclaw/skills/tree/main/skills/jmanhype/wreckit-skill

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Wreckit Skill (jmanhype 分支) 应用场景

• 在隔离、安全的微型虚拟机中运行自主开发循环，防止主机污染。
• 通过自主构思（dreaming）自动发现技术债并生成路线图。
• 通过并行处理多个开发项来扩展软件生产规模。
• 使用诊断和遗传算法优化工具自愈并修复损坏的代码库。
• 生成功能可视化视频并从已完成的任务中提取可重用的编码模式。

1. 用户通过 Clawdbot 界面触发命令，如 "Wreckit run" 或 "Wreckit dream"。
2. 该技能调用 run-wreckit.mjs 脚本，该脚本与全局安装的 Wreckit CLI 进行通信。
3. 根据配置，系统会启动一个瞬时 Sprite VM（Cattle 模式）或直接在主机上执行（Pet 模式）。
4. 智能体处理分配的项目 ID，在隔离环境中执行代码更改、测试或诊断。
5. 完成后，VM 将被销毁，技能会根据需要处理状态更新、模式提取或自动回滚。

Wreckit Skill (jmanhype 分支) 配置指南

要使用此技能，请确保已全局安装 jmanhype 分支的 Wreckit CLI，并且可以在环境路径中访问。

# 从 jmanhype 分支安装 Wreckit CLI
npm install -g @jmanhype/wreckit

# 在您的智能体环境中验证安装
wreckit --version

安装完成后，您可以在智能体界面中使用指定的触发短语与之交互，以管理您的 Openclaw Skills 工作流。

Wreckit Skill (jmanhype 分支) 数据架构与分类体系

Wreckit 使用以项目、阶段和提取的智能为核心的结构化分类法来组织其数据。

Wreckit Skill (jmanhype Fork)

First time? Read SETUP.md first to install the Wreckit CLI from the jmanhype fork.

Connects Clawdbot to the jmanhype fork of Wreckit, the Autonomic Software Factory. This version is optimized for Cattle Architecture (Ephemeral Sprite VMs) and high-concurrency loops.

Script: skills/wreckit/scripts/run-wreckit.mjs Engine: wreckit (Global CLI from jmanhype/wreckit)

Trigger Phrases

When human says:

The "Cattle" Advantage (Default)

Unlike standard agent loops, this version of Wreckit uses Ephemeral Sprite VMs. Every task spins up a fresh Firecracker microVM, executes code in isolation, and vanishes. No state contamination, maximum security.

The "Pet" Mode (Local)

If you need raw speed or are debugging locally, say "mode pet". This runs the RLM agent directly on the host machine.

Usage Guide

1. Checking Status

User: "Wreckit status" Bot: Lists all active items (Idea, Planned, In Progress, Done).

2. Running a Task

User: "Wreckit run 096" Bot: "Starting Wreckit run for item 096... [Streams logs] ... Done."

3. Dreaming (Autonomy)

User: "Wreckit dream" Bot: Scans your codebase for technical debt and creates new items automatically.

4. Rollback

User: "Wreckit rollback 096" Bot: Reverts item 096 to its pre-merge state using git operations.

With force: "Wreckit rollback 096 force" - Forces rollback even if item is not in 'done' state.

5. Next Item

User: "Wreckit next" Bot: Processes the single next runnable item with dependency satisfaction.

6. Learn (Pattern Extraction)

User: "Wreckit learn all" Bot: Extracts patterns from all completed items and compiles them into skills.

Specific item: "Wreckit learn item 096" By phase: "Wreckit learn phase done"

7. Summarize (Video Generation)

User: "Wreckit summarize all" Bot: Generates 30-second feature visualization videos for all completed items.

Specific item: "Wreckit summarize item 096" By phase: "Wreckit summarize phase done"

8. Geneticist (Prompt Optimization)

User: "Wreckit geneticist" Bot: Analyzes healing logs from the last 48 hours to identify and fix recurrent error patterns.

Custom window: "Wreckit geneticist time-window 24 min-errors 5"

Advanced Flags

Global Flags (apply to all commands)

• --cwd - Override the working directory
• --parallel - Process N items in parallel (default: 1)
• --verbose - Enable verbose output
• --dry-run - Show what would be done without making changes

Mode Selection

• --mode cattle (default) - Use ephemeral Sprite VMs
• --mode pet - Run locally without Sprites

Command-Specific Flags

Rollback

• --force - Force rollback even if item is not in 'done' state

Learn

• --item - Extract patterns from specific item
• --phase - Extract patterns from items in specific phase state
• --all - Extract patterns from all completed items
• --output - Output path for skills.json (default: .wreckit/skills.json)
• --merge - Merge strategy: append|replace|ask (default: append)
• --review - Review extracted skills before saving

Summarize

• --item - Generate video for specific item
• --phase - Generate videos for items in specific state
• --all - Generate videos for all completed items

Geneticist

• --auto-merge - Automatically submit PRs for optimized prompts
• --time-window - Analyze healing logs from last N hours (default: 48)
• --min-errors - Threshold for recurrent pattern detection (default: 3)

Configuration

This skill assumes wreckit is installed globally or available in the path. If running in Docker, ensure the container has access to the project directory.

Parallel Processing

When using --parallel flag with commands that process multiple items (like learn --all or summarize --all), the wrapper will process N items concurrently. For bot safety, default is 1 (sequential processing).

Working Directory Override

Use --cwd to specify a different project directory than the current working directory. This is useful for managing multiple projects or operating in a containerized environment.