Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Code review agent for embedded and firmware projects. Runs Claude Code and Codex independently against the same diff via ACP, then cross-compares findings to surface issues either model alone would miss. Covers memory safety, interrupt hazards, RTOS pitfalls, hardware interface bugs, and C/C++ anti-patterns across bare-metal MCU, FreeRTOS, Zephyr, ThreadX, and Linux embedded targets.
Running two independent models against the same diff and cross-comparing their findings catches blind spots that any single model or human reviewer would miss.
When to use it
- Reviewing a FreeRTOS patch before merging to a release branch
- Catching ISR race conditions in NFC firmware changes
- Auditing DMA buffer handling in a new MCU driver
- Checking a GitHub PR for a bare-metal bootloader update
- Finding undefined behavior in C firmware before a production build
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: /embedded-review ~/Documents/dec/firmware-pro2 HEAD~5..HEAD
- 1Runs prepare-diff.sh to extract the git diff, MCU/RTOS target info, and changed file stats
- 2Assesses diff size and detects critical paths (ISR, DMA, NFC) to select dual-model mode
- 3Spawns parallel ACP sessions for Claude Code and Codex with independent review prompts
- 4Cross-compares findings, classifying each as consensus, Claude-only, Codex-only, or contradicting
- 5Outputs structured P0-P3 severity report and prompts user to choose a next-steps action
Structured review report with severity-ranked findings, a cross-review analysis table showing consensus vs. model-specific catches, and a menu to fix all, fix P0/P1 only, or target specific issues