The Agent Skills Directory

[DATA_EXFILTRATION]: The skill packages and transmits the contents of a specified directory to Vercel's deployment service. While the destination is a well-known and trusted service provider, the script's packaging logic (using tar) only excludes .git and node_modules. Consequently, other sensitive files such as .env, .npmrc, or secret configuration files located in the project root will be included in the upload unless manually removed or placed in excluded directories.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data to perform framework detection and file management.
Ingestion points: Reads and parses package.json and scans for .html files within the user-specified project directory in deploy.sh.
Boundary markers: None; the script treats project file contents as trusted input for its logic.
Capability inventory: Includes file system write access (mv for renaming HTML files), file system read access (tar, cat), and network transmission (curl).
Sanitization: None; the script uses basic string matching via grep on ingested file contents to identify framework dependencies.
[COMMAND_EXECUTION]: The script utilizes standard shell utilities (tar, find, mv, curl) to archive project data and transmit it to the deployment API.

vercel-deploy