ClawHub

ClawHub is the centralized distribution point for OpenClaw agent skills: packaged, shareable bundles of behavior that extend what an agent can do, covering things like code review, file triage, or web research. A developer publishes a skill to ClawHub; any OpenClaw user can install it via the skills.sh installer or Claude Code's plugin marketplace. Think of it like npm for agent behavior, with all the supply-chain exposure that analogy implies.

In February 2026, security researchers found that a significant portion of the ClawHub catalog contained malicious code. Firm Koi Security flagged roughly 12% of scanned skills as harmful, while SlowMist issued high-severity alerts for hundreds more. The attack patterns included prompt injection payloads, backdoors, exposed API keys, and keyloggers, all packaged inside skills that looked legitimate. Because skills run with the same permissions as the agent itself, including access to files, the shell, and credentials, a poisoned skill is a serious threat.

OpenClaw responded by partnering with VirusTotal to scan every published skill for known threats and code anomalies. The episode established ClawHub as the canonical example of the agent skill supply-chain problem: the same open ecosystem that makes skill sharing fast and composable also makes it a distribution vector for attackers. Before installing any third-party skill, builders should manually check the SKILL.md, review declared dependencies, and be cautious of permissions the skill requests.