mcpField
L
lazycat-aipod-developer

lazycat-aipod-developer

github.com/whoamihappyhackingCloud & DevOps59161 installsUpdated 3w ago
Repository

About lazycat-aipod-developer

A skill for building AI applications deployable to the Lazycat AI Pod. It covers deployment constraints such as Jetson GPU defaults, Traefik routing, and persistent data paths, and points to the full packaging specification.

Key features

  1. 01AI Pod app development guidance
  2. 02Jetson GPU container constraints
  3. 03Traefik routing requirements
  4. 04Data and cache path conventions
  5. 05Reference to packaging specifications

Use cases

  1. 01Building AI apps for the AI Pod
  2. 02Configuring Traefik routes for app services
  3. 03Setting persistent data and cache paths

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/whoamihappyhacking/lazycat-skills --skill lazycat-aipod-developer

After installing, type/lazycat-aipod-developerin Claude Code to confirm it loaded.

Bundled filesreferences/aipod-app-spec.md

Details

From the project README

懒猫 AI 算力舱应用构建指南

你现在是懒猫微服 AI 算力舱(AI Pod)的应用开发专家。本技能聚焦于 如何构建一个可部署到算力舱的 AI 应用。

核心概念

  • AI 应用 = 微服应用 + 算力舱 AI 服务(和/或 AI 浏览器插件)
  • 安装到微服后,会自动将 AI 服务部署到算力舱
  • 算力舱基于 NVIDIA Jetson,Docker 默认使用 nvidia-runtime,容器内直接可用 GPU,无需显式配置
  • 服务网关使用 Traefik,通过 Host 规则转发,域名必须以 -ai 结尾

行动指令

当用户需要构建 AI 算力舱应用时,请读取 references/aipod-app-spec.md 获取完整的打包规范、docker-compose 编写要求、Traefik 路由配置、环境变量说明、浏览器插件打包、进度提示集成、应用发布等详细内容。


给 AI 引擎的强制约束:

  1. 算力舱 Docker 默认使用 nvidia-runtime,不要在 docker-compose 中添加 gpus 或 runtime: nvidia 配置
  2. Traefik 的 Host 域名必须以 -ai 结尾,否则无法转发
  3. 服务必须加入 traefik-shared-network 网络才能被 Traefik 接管
  4. 使用 LZC_AGENT_DATA_DIR 做数据持久化,LZC_AGENT_CACHE_DIR 做缓存,LZC_SERVICE_ID 做路由命名
  5. 当需要获取微服名称时,执行 lzc-cli box default,不要询问用户

Comments

Markdown supported