computer-use-agents
Warn
Audited by Socket on Oct 31, 2026
1 alert found:
SecuritySecurityreferences/patterns.md
MEDIUMSecurityMEDIUM
references/patterns.md
This fragment provides computer-use and browser-agent examples and shows no clear malicious payload or unauthorized exfiltration destination. It does send screenshots and task context to the Anthropic API, executes model-directed actions, and includes risky shell execution, unrestricted editor paths, shell-interpolated Docker arguments, and potentially exposed VNC/API ports. Treat the examples as security-sensitive and do not deploy them without strict command, path, network, and access controls; severity is high for an exposed or untrusted deployment, but malicious intent is not indicated.
Confidence: 98%Severity: 78%
Audit Metadata