ai-agents 3 OpenClaw Security in February 2026: One-Click RCE, 40K Exposed Instances, and a Marketplace Problem Feb 15, 2026 The Read-Only Triage Agent Pattern: Architectural Defense Against Prompt Injection Feb 15, 2026 OpenClaw Subagents: Context Isolation, Workspaces, and Security Lockdown Feb 15, 2026