BrokenClaw - RCE in OpenClaw via Gmail Hook (Security Research)

Security research documenting 0-click RCE in OpenClaw via Gmail pub/sub hook. Attack chain: malicious email → prompt injection → agent clones malicious .openclaw repo → gateway restart → malicious plugin executes → reverse shell. Only affects users with full Gmail pub/sub webhook enabled (not gogcli users). Tested on versions 2026.2.2-3 through 2026.2.21-2 with both GPT5.2 and Opus4.6. Why we added it: Critical security advisory — if you use Gmail webhook, you need to read this.