RDK Device Knowledge Skills · RDK 设备知识技能

Source-backed knowledge for D-Robotics RDK boards (and Jetson / Raspberry Pi / Rockchip), written in the official Claude skill format. Any skill-aware host (Claude Code, Moss, RDK Studio) scans skills/** directly.

面向 D-Robotics RDK(及 Jetson / 树莓派 / Rockchip)的 source-backed 知识,采用 Claude 官方 skill 格式。任意支持 skill 的宿主(Claude Code、Moss、RDK Studio)可直接扫描 skills/** 加载。

Skills · 技能清单

Skill	What it covers · 用途
rdk-ecosystem	Board selection & cross-platform comparison · 选型 / 跨平台对比 / 能否跑某模型
rdk-hardware	Per-board hardware facts (specs, 40PIN, CAN, power, IP) · 板型硬件事实(规格 / 40PIN / CAN / 供电 / IP)
rdk-board-knowledge	Board ID, error diagnosis, FAQ, S-series flashing · 板型识别 / 报错诊断 / FAQ / S 系烧录
rdk-system-config	System config (network, freq, thermal, autostart) · 系统配置(网络 / 调频 / 温控 / 自启)
rdk-device	Deploy your own model end-to-end + first boot + camera · 自训模型端到端部署 + 开箱 + 摄像头
rdk-model-zoo	Ready-made precompiled BPU models · 取现成预编译 BPU 模型
rdk-ros	TROS/ROS2 env, nodes, perception · TROS/ROS2 环境 / 节点 / 感知
rdk-multimedia	Hardware codec & camera pipeline · 硬件编解码与相机流水线
rdk-peripheral-cookbook	Drive peripherals (GPIO/I2C/SPI/CAN/audio) · 驱动外设(GPIO/I2C/SPI/CAN/音频)
rdk-accessories	Official accessories (cameras, IMU, expansion boards) · 官方配件(相机 / IMU / 扩展板)
rdk-llm-deployment	On-board LLM/VLM chat + voice stack · 端侧 LLM/VLM 对话 + 语音栈
rdk-embodied-lerobot	Embodied policy deploy (ACT / VLA / Pi0) · 具身策略部署(ACT / VLA / Pi0)
rdk-board-delegate	S-series heterogeneous dev (MCU/hbmem/EtherCAT) + OpenClaw delegate · S 系异构开发 + OpenClaw 委派
rdk-command-manual	RDK command syntax / flags / boards · RDK 命令语法 / 选项 / 适用板型
rdk-doc-finder	Locate the official doc-site chapter & URL · 定位官方文档站章节与 URL
rdk-source-map	Navigate the D-Robotics GitHub org repos · 在 D-Robotics GitHub 组织里定位仓库
jetson-knowledge	NVIDIA Jetson ecosystem & toolchain · NVIDIA Jetson 生态与工具链
rpi-knowledge	Raspberry Pi ecosystem & GPIO · 树莓派生态与 GPIO
rk-knowledge	Rockchip RK3588 NPU & RKNN · Rockchip RK3588 NPU 与 RKNN
host-software-dev	RDK Studio desktop client dev · RDK Studio 桌面客户端开发

Vendored official skills · 内置官方 skill

General-purpose tooling skills copied unmodified from anthropics/skills (Apache-2.0), kept here for maintainers. See THIRD_PARTY_NOTICES.md.

从 anthropics/skills 原样收录的通用工具 skill(Apache-2.0),供维护者使用,详见 THIRD_PARTY_NOTICES.md。

Skill	What it's for · 用途
skill-creator	Create / improve / measure skills · 创建 / 改进 / 评测 skill(维护本仓用)
mcp-builder	Build MCP servers · 构建 MCP server(把知识做成 MCP 服务时用)

Add a skill · 新增 skill

A skill is one folder under skills/<name>/. Follow the shape of rdk-device as the reference template.

一个 skill 就是 skills/<name>/ 下的一个文件夹。以 rdk-device 为参考模板。

skills/<name>/
├── SKILL.md           # required · 必需
├── references/*.md     # detailed tables/notes · 详细查表(可选)
├── evals/evals.json    # test prompts + assertions · 测试集(可选)
├── scripts/            # deterministic helpers · 确定性脚本(可选)
└── assets/             # templates/config · 模板/配置(可选)

EN

1. SKILL.md frontmatter — name (must equal the folder name) and a description that states what it does + when to use (a bit pushy to avoid under-triggering), includes Chinese 触发词 keywords, and routes to sibling skills.
2. Body — write in English, imperative and scannable: a decision table where useful, numbered workflows, 3–4 worked examples (real question → how to answer), a do/don't pitfalls table, and a reference map.
3. Bundled resources — put long detail in references/*.md (English; add a table of contents if a file exceeds 300 lines). Add evals/evals.json. Add scripts/ or assets/ only when genuinely useful.
4. Stay source-backed — cite official docs (D-Robotics / NVIDIA / Raspberry Pi / Rockchip); don't invent facts.
5. Validate — run node scripts/validate-skills.mjs (must pass: name == folder, every references/ link resolves).

中文

1. SKILL.md frontmatter —— name(必须等于文件夹名)+ description:写清做什么 + 何时用(稍微"主动"些以免漏触发),带中文 触发词 关键词,并路由到相邻 skill。
2. 正文 —— 用英文写,imperative、可扫读:需要时给决策表、编号 workflow、3–4 个 worked example(真实问题 → 怎么答)、do/don't 速查表、reference map。
3. 附带资源 —— 长内容放 references/*.md(英文;超 300 行加目录)。加 evals/evals.json。scripts/、assets/ 仅在确有用处时加。
4. 保持 source-backed —— 引用官方文档(D-Robotics / NVIDIA / 树莓派 / Rockchip),不臆造事实。
5. 校验 —— 运行 node scripts/validate-skills.mjs(必须通过:name == 文件夹名、每个 references/ 链接都存在)。

License

MIT