razyos/firmware-ai-reviewer

firmware-ai-reviewer is a deterministic, prompt-chained static analysis tool for embedded C firmware (TI CC2652R7 Cortex‑M4F and STM32 F4/F7/H7 Cortex‑M4/M7, FreeRTOS/HAL). It addresses embedded-specific failures of generic LLM reviews (context dilution, constraint conflicts, lack of verifiability). Key features: - Multi‑phase pipeline: Phase 0 injects local headers (line numbers preserved); Phase 1 routes code to domain tags (RTOS, ISR, DMA, etc.); Phase 2 assembles platform-aware expert prompts; Phase 3 runs parallel experts (RTOS/ISR, Memory/Pointer, Hardware/DMA, UART, Security); Phase 4 merges and de‑dupes findings. - Each expert gets only domain rules, must produce a reasoning_scratchpad, and emits typed JSON (line_number, severity, rule, description, fix). - Deterministic inference (temperature=0.0), eval suite with seeded bugs and ground-truth rule IDs, configurable model profiles and CLI for single-file review or full eval.