Dimensions overall score 4.0

GitHub Code Pulse

No public code linked for this paper yet.

• Users of computer-use agents often lack concrete mental models of what skills can do, what resources agents can access, and what changes may remain after execution or removal.result 85%
• There is a significant gap in user understanding regarding the authority delegated to computer-use agents, what the agent actually did during task execution, and whether the system has been safely removed.limitation 90%
• The OpenClaw ecosystem has been associated with security incidents including credential leaks and prompt injection vulnerabilities.result 95%
• A traceability framework (AgentTrace) can improve understanding of agent behavior, support anomaly detection, and foster more calibrated trust.result 80%
• Personalized computer-use agents differ from conventional chatbots by being able to install skills, invoke tools, access private resources, and modify local environments.technical 95%
• The problem of understanding agent actions is a combined issue of risk understanding and post-hoc auditability.technical 90%
• Users often need procedural understanding of what an AI agent did before they can make safety or trust judgments.result 85%
• The research methodology involved building a multi-source corpus of the OpenClaw ecosystem and conducting an interview study with users and practitioners.method 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card