LLMs Confidently Lie About Your Microcontroller
Embedded development is the field where the soldering iron and the IDE sit closest together. Naturally, AI has arrived here too: ask a model to write firmware and it produces tidy code. It compiles on the first try and flashes without errors. And then the screen shows colorful noise. The code isn't the problem. The display uses an ST7789 controller designed for a 240×320 panel, but the glass on…
In the realm of embedded development, AI has found its way into the process. When prompted to write firmware, models produce tidy code that compiles and flashes without errors. However, the results often display colorful noise, indicating a problem. The issue lies not with the AI's intelligence but in the scattered nature of knowledge about specific boards. This knowledge is scattered across datasheets, manuals, and debugging experiences, leaving AI without the necessary context to avoid issues.
For example, the Waveshare ESP32-C6-LCD-1.47 display controller expects a 34-pixel offset to prevent the image from shifting sideways. Similarly, the Raspberry Pi Pico's GPIO23, 24, 25, and 29, while programmed as real GPIOs, lack physical presence on the 40-pin header. The ESP32's ADC2 is fully occupied by Wi-Fi, causing analog sensor circuits using ADC2 to fail silently.
The STM32F411 microcontroller's USB interface fails to function at 100 MHz, and HAL_Delay() causes the board to hang due to manual definition of SysTick_Handler, a detail rarely documented.
To address these challenges, an open collection of skills for Claude Code has been created, one skill per board. Each skill is a directory containing a skill summary, in-depth details, and a working project template, extracted from real-world debugging experiences. This approach aims to fill the gaps in AI's knowledge about specific boards, such as pin mappings, chip characteristics, and potential pitfalls.
The project currently covers various microcontroller families, including STM32, ESP32, Arduino Nano, and Raspberry Pi RP2040. By cloning the repository and installing the skills, users can automatically load the necessary knowledge for their boards when working in Claude Code. This allows the AI to avoid costly mistakes and provides a persistent knowledge base across sessions, preventing the "forgetting" issue that arises during long-running projects.
By versioning the skills alongside the code, teams can benefit from shared knowledge and collaborate more effectively.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
