vercel-ai-sdk

This example implements resumable AI chat streams and shows no evidence of malware or intentional exfiltration to an external destination. The principal security concern is that the shown create and resume endpoints do not authenticate or authorize chat IDs, potentially allowing unauthorized chat modification or stream disclosure; input validation is also absent. Severity is moderate and depends on the omitted storage and application middleware. No malicious domains, URLs, IP addresses, or credentials are present in the code.

This documentation demonstrates AI-controlled desktop, file-editing, and shell tools; the bash example executes model-supplied commands through execSync, creating a high-impact command-execution risk if run outside an isolated environment. The computer and editor implementations are omitted, so their exact effects cannot be assessed. No evidence of intentional malware, credential theft, or exfiltration to a suspicious domain appears in this fragment; the only named destination is vercel[.]com. Severity is high if exposed to untrusted model inputs or run with meaningful host privileges, and the notes recommend isolation and least privilege.