vercel-deploy

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The fallback deploy script intentionally uploads the user's project tarball to an external endpoint (codex-deploy-skills.vercel.sh) and the skill instructs running it (and escalating network access) in sandbox contexts, which constitutes deliberate data exfiltration risk and potential unauthorized deployment/claiming of user projects.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The fallback deploy script posts the user's project tarball at runtime to "https://codex-deploy-skills.vercel.sh/api/deploy", which performs the remote deployment (executes remote code) and the skill relies on that endpoint for no-auth deployments.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs the agent to request escalated sandbox/network permissions (sandbox_permissions=require_escalated) to bypass sandboxing for deploy commands, which directs the agent to bypass security controls on the host.