OpenKits Skills: Ensine IA a Conhecer Sua Placa de Desenvolvimento
jlckfb/openkits-skills
IA entende sua placa: pinos, periféricos e workflow de compilação automáticos
Ver no GitHubColeção de skills para ensinar agentes de IA (Claude, Codex, Cursor e 28+ outros) a entender pinos, periféricos e toolchain de placas de desenvolvimento LCSC. Permite criar, desenvolver, compilar e gravar projetos inteiros usando linguagem natural, sem conflitos de pinos ou configuração manual de periféricos.
Instalação
npx convez-ai jlckfb/openkits-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar projetos de firmware com descrição em linguagem natural
- Evitar conflitos de pinos em placas de desenvolvimento
- Automatizar scaffold de projetos com templates corretos do SDK
- Acelerar ciclo completo engenharia (design → código → compilação → gravação)
- Documentar e compartilhar conhecimento técnico de placas entre engenheiros
Veredito

Ferramenta especializada que reduz drasticamente tempo de onboarding em novos boards ao ensinar IA a lidar com complexidades técnicas (pin mapping, clock trees, exemplos do SDK). Execução totalmente local (sem cloud), ideal para equipes de hardware/firmware que usam placas LCSC com agentes de código.
README
A tradução para o português deste README ainda está na fila.
openkits-skills: Teach AI to Know Your LCSC Dev Board
中文 · by LCSC Development Board
AI agent skill collection for LCSC development boards. Teach Claude Code, Codex, Trae, Cursor, and 28+ other AI coding agents to understand your board's pins, peripherals, and toolchain.
What Is This
Every dev board has its own pin assignments, peripheral configurations, and toolchain rules. This knowledge is scattered across schematics, datasheets, and SDK docs — engineers have to dig through it every time they pick up a new board. openkits-skills packages this knowledge into structured skill files that AI agents can read directly: which pins are taken, where the SDK examples are, how to build and flash.
You describe what you want. The AI handles the rest.
What Problems It Solves
The most time-consuming part of starting with a new board is rarely the business logic:
- Pin conflicts: On-board peripherals already occupy a bunch of pins. Picking one at random often causes silent hardware failures — no error, the hardware just doesn't work
- Peripheral configuration: Clock trees, pin mux, peripheral instances — each has its own pitfalls and scattered docs
- Finding examples: The SDK
examples/directory has dozens of subdirectories; finding the right one takes time - Cryptic build errors: Generated macro names come from the toolchain, not your code — guessing them wrong means repeated trial and error
openkits-skills captures this board-level knowledge in skill files so the AI learns it once, and every engineer using that board avoids the same traps.
How It Works
You describe what you want (natural language)
│
▼
AI agent (Claude Code / Codex / Trae ...)
│ reads skill files
▼
openkits skill
├── Pin occupation table (which pins are off-limits)
├── Peripheral reference docs (UART / GPIO / PWM / I2C / SPI / ADC ...)
└── Workflow rules (Think → Plan → Code → Verify)
│ calls local scripts
▼
scaffold.py ──→ generate project (.syscfg + source + project config)
│
▼
build.py ──→ toolchain compile ──→ firmware
│
▼
flash.py ──→ flash to board
The entire chain runs locally — no code is uploaded, no cloud build required.
Available Skills
TI MSPM0 Platform
| Skill | Type | Description |
|---|---|---|
| mspm0-ccs | Platform | MSPM0 SysConfig/DriverLib general rules |
| mspm0kit-tianqiaoxing | Board | Tianqiaoxing MSPM0G3519 dev board |
| mspm0kit-tianmengxing | Board | Tianmengxing MSPM0G3507 dev board |
Other Platforms
Coming soon: STM32 / ESP32 / GD32 / Rockchip / K230 / RP2350 / CW32 / Renesas / FPGA / SF32LB52
Example (Tianmengxing MSPM0G3507)
You: Create a UART0 debug project, 115200 baud
AI: Here's what I'll create:
- Project: uart_debug
- Template: uart_rw_multibyte_fifo_poll (SDK official example)
- Pins: PA10(TX) / PA11(RX), connected to on-board CH340
- Clock: 80 MHz HFXT
Confirm to proceed?
You: Confirm
AI: [scaffold.py uart_debug uart_rw_multibyte_fifo_poll]
[cleanup.py → build.py]
Build succeeded: uart_debug/Debug/uart_debug.out
Flash: python scripts/flash.py uart_debug
Install
Recommended: npx skills (28+ agents)
# Install a single board skill
npx skills add lckfb/openkits-skills -s mspm0kit-tianmengxing -a claude-code
# Install all skills for a platform
npx skills add lckfb/openkits-skills -s mspm0-ccs mspm0kit-tianmengxing -a claude-code
# Target multiple agents at once
npx skills add lckfb/openkits-skills -s mspm0kit-tianmengxing -a claude-code codex trae
China mirror (Gitee):
# Tianmengxing
npx skills add https://gitee.com/lcsc/openkits-skills.git -s mspm0kit-tianmengxing -a claude-code
# Tianqiaoxing
npx skills add https://gitee.com/lcsc/openkits-skills.git -s mspm0kit-tianqiaoxing -a claude-code
# MSPM0 platform skill
npx skills add https://gitee.com/lcsc/openkits-skills.git -s mspm0-ccs -a claude-code
Select Global when prompted, then enter yes to confirm.
Manual Install
Copy skills/<skill-name>/ to the agent's skills directory:
| Agent | Install Path |
|---|---|
| Claude Code | ~/.claude/skills/<name>/ |
| Codex | ~/.agents/skills/<name>/ |
| Trae | ~/.trae/skills/<name>/ |
| Cursor | ~/.cursor/skills/<name>/ |
Toolchain Path Setup (MSPM0 only)
MSPM0 skills need to know where CCS and the SDK are installed. Run once on first use:
# Tianmengxing
python ~/.claude/skills/mspm0kit-tianmengxing/scripts/setup.py
# Tianqiaoxing
python ~/.claude/skills/mspm0kit-tianqiaoxing/scripts/setup.py
Follow the prompts (press Enter to accept defaults). Settings are saved to config.json and don't need to be repeated. Other platform skills don't require this step.
Uninstall
npx skills
# List installed skills
npx skills list
# Remove a specific skill
npx skills remove mspm0kit-tianmengxing
# Remove global install
npx skills remove mspm0kit-tianmengxing -g
# Remove all
npx skills remove --all
Manual
rm -rf ~/.claude/skills/mspm0kit-tianmengxing # Claude Code
rm -rf ~/.agents/skills/mspm0kit-tianmengxing # Codex
Add a New Board
- Identify the MCU platform. If no platform skill exists, create one first (see CLAUDE.md).
- Copy
templates/new-board/, fill in the 3 template sheets, and create the skill directory underskills/:
skills/<board-name>/
├── SKILL.md # name, requires, pin table, workflow
├── scripts/ # scaffold.py, build.py, flash.py
├── peripherals/ # one .md per peripheral (gpio.md, uart.md, …)
└── examples/ # board-specific examples (with manifest.json)
- Update the "Available Skills" table above.
Full spec: CLAUDE.md.