Alice Launches Caterpillar to Police AI Agent Skills in OpenClaw

New updates have been reported about Alice.

Alice introduced Caterpillar, a free open-source scanner built for OpenClaw that inspects agent skills for prompt injection pathways, unsafe tool permissions, and obfuscated logic, leveraging the company’s RabbitHole threat intelligence for adversarial context. The tool already prevented misuse by catching several malicious skills across a user base exceeding 6,000, underscoring the operational risks emerging as agent ecosystems shift from experimentation to production workloads. CEO Noam Schwartz said the release targets a widening security gap as enterprises load increasingly autonomous skills that deliver behavior rather than simple features, positioning Caterpillar as an early-warning system to avoid costly incidents. Caterpillar’s open-source availability should accelerate community-led auditing and extensions, buttressing Alice’s strategy to embed its safety stack deeper into AI agent supply chains. The product follows the firm’s broader mandate to serve as the trust and security layer for GenAI builders, adding runtime guardrails to its existing model-hardening and red-teaming portfolio. Executives should note that the tool can be operationalized without licensing cost and feeds into Alice’s telemetry-driven intelligence, potentially reducing vendor sprawl for organizations already using RabbitHole insights.