Briefing
Every hardware call is checked against the bench configuration, run with timeouts, logged and answered with structured JSON that lists likely causes and the report path. Flashing is refused while raw debugger commands or mass erase are allowed. Over MCP an agent can narrow its own permissions but not widen them, and each action is leased machine-wide and written to a SHA-256 audit chain. YAML test plans flash, reset, write and read with comparators.
Support is defined by debug probe rather than board: ST-Link through OpenOCD or STM32CubeProgrammer, and CMSIS-DAP probes through pyOCD. CAN runs over PCAN, SocketCAN or a custom bridge. The README embeds a video of an agent setting up a bench and reading Hello World back from a board; we have not watched it.
Written by the Boards & Bots curator