agent-s
Warn
Audited by Socket on Nov 6, 2026
1 alert found:
SecuritySecurityreferences/upstream.md
MEDIUMSecurityMEDIUM
references/upstream.md
The supplied fragment is documentation summarizing an audit, not executable code, and contains no direct evidence of malware. It reports that Agent S sends desktop screenshots to configured model endpoints and executes model-generated Python, with optional arbitrary local Python and Bash execution; the reported absence of per-action confirmation creates significant operational risk. No fixed suspicious domain, IP address, or credential value is present. Severity is high for use with untrusted models or sensitive desktops, but malicious intent is not established by this fragment.
Confidence: 94%Severity: 72%
Audit Metadata