Security risk report for @forgecat/openai_skills_vercel-deploy v0.0.11
Source Integrity
Low
Profile is attributed to OpenAI with a public GitHub repository link and specific commit hash, enabling verification.
MIT license and clear provenance metadata support supply-chain transparency.
Agent Intent
Low
Content describes legitimate deployment workflow (check CLI, run deploy, show URL) without injecting hidden instructions or role hijacking.
No guidance to exfiltrate credentials, read sensitive files, or hide instructions; fallback script is documented as a standard deployment helper.
Troubleshooting section advises requesting user consent before escalating permissions—transparent, not deceptive.
Details
Evidence
"Always deploy as preview (not production) unless the user explicitly asks for production."
"Tell the user: Your deployment is ready at [previewUrl]. Claim it at [claimUrl] to manage your deployment."
"Do not curl or fetch the deployed URL to verify it works. Just return the link."
Permissions
PRM-000Medium
The skill instructs the agent to execute shell commands (`command -v vercel`, `vercel deploy`, `bash ...deploy.sh`) which fall into the shell/file-execution category.
Shell execution is broadly consistent with a deployment workflow, but running arbitrary bash scripts from a skill-bundled path introduces moderate authority beyond a purely advisory role.
The request for escalated sandbox permissions (`sandbox_permissions=require_escalated`) for network access is a notable elevation, though it is contextually justified for deployment.
Details
Evidence
"bash \"$skill_dir/scripts/deploy.sh\""
"vercel deploy [path] --prod -y"
"sandbox_permissions=require_escalated"
MCP Risk
Low
No MCP servers are declared in this profile.
Deployment is handled via standard Vercel CLI and documented shell scripts, not arbitrary binary execution or hidden tool definitions.
ForgeCat