OpenClaw 3.31 Security Audit: What the Obfuscation Detection Patch Reveals About Self-Hosted Trust Models
Analysis
Hands-On: The Obfuscation Detection Override That Breaks Legitimate Agent Exec
Solomon Neas documents a real production issue where 3.31's hardcoded obfuscation detection ignores ask:off and blocks normal agent commands like base64 payloads and long one-liners. This is the most actionable writeup for self-hosted operators running AI agents with exec access.
Release
OpenClaw v2026.3.31 Official Changelog: Six Breaking Changes and the Full Security Surface
The official release notes detail every breaking change including gateway auth strictness, node command pairing requirements, and plugin install fail-closed behavior. Required reading before you update any production deployment.
4.2 moves xAI and Firecrawl config into plugin-owned namespaces and centralizes provider HTTP auth and routing across every transport path. The exec environment also gets additional pivot-prevention for package roots, language runtimes, and credential locations that were still exploitable after 3.31.
Security
Plugin Installs Now Have a Bouncer: The ClawHub Malware Context Behind 3.31 Security Changes
This post connects the install-time dangerous-code scanner directly to the Cisco research finding 26% of audited skills contained vulnerabilities and 800+ confirmed malicious entries on ClawHub. The security hardening in 3.31 was a direct response to a documented ecosystem threat, not just defensive tidying.
News
Operator Briefing: Background Task Control Plane, QQ Bot, and Native Codex Search in 3.31
The clearest plain-language summary of 3.31's three major themes: the SQLite-backed unified task ledger, QQ Bot as a first-party bundled channel, and native Codex web search with managed-tool suppression. Good starting point if you want the operator summary before digging into the full changelog.
Stay Ahead
Delivered each morning.