The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains deceptive metadata that attempts to impersonate an official vendor. The YAML frontmatter incorrectly lists the author as 'vercel', contradicting the actual platform-identified author 'Mrziadd0'. This deception can cause users to misjudge the safety and authenticity of the skill's operations.
[DATA_EXFILTRATION]: The fallback scripts (deploy.sh and deploy-codex.sh) perform automated data exfiltration by archiving the project's source code and uploading it to remote endpoints (https://claude-skills-deploy.vercel.com/api/deploy and https://codex-deploy-skills.vercel.sh/api/deploy). While the scripts exclude .env files, the core logic and proprietary source code are still transmitted to an unverified external service.
[COMMAND_EXECUTION]: The skill executes multiple local shell scripts and complex CLI commands with user-supplied arguments, providing a broad attack surface for command injection if project paths or team slugs are manipulated.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to install external software globally (npm install -g vercel), which introduces dependencies from a public registry into the user's environment.
[DATA_EXFILTRATION]: Indirect Prompt Injection Surface.
Ingestion points: Project configuration files like package.json are read and parsed via shell commands (cat, grep).
Boundary markers: No delimiters or safety warnings are used to prevent the agent from interpreting content within these project files as instructions.
Capability inventory: The skill possesses extensive capabilities including file system read/write, network access (uploads), and software installation.
Sanitization: There is no validation or sanitization of the project files before they are archived and exfiltrated to the remote deployment service.

deploy-to-vercel