The Agent Skills Directory

[METADATA_POISONING]: The skill metadata falsely identifies the author as 'vercel', which is a well-known and trusted organization. This impersonation can lead users or agents to accord the skill undue trust. Evidence: 'author: vercel' in the YAML frontmatter of SKILL.md.
[CREDENTIALS_UNSAFE]: The skill instructs the agent to search local .env files and environment variables for sensitive Vercel access tokens. While necessary for functionality, this involves scanning potentially sensitive files for secrets.
Evidence: Instructions to use grep '^VERCEL_TOKEN=' .env and grep -i 'vercel' .env in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill requires the global installation of the official Vercel CLI from the public npm registry.
Evidence: npm install -g vercel command in SKILL.md.
[COMMAND_EXECUTION]: The skill relies on executing various shell commands to perform its tasks, including file searching, environment variable manipulation, and CLI-based deployment.
Evidence: Usage of grep, export, npm, git, and vercel commands throughout SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrustworthy external data (such as project URLs and .env file content) and uses it to drive sensitive operations.
Ingestion points: PROJECT_URL variable, .env file content, and git remote output.
Boundary markers: Absent; there are no specific delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill can perform global package installation, git pushes, and deployment operations.
Sanitization: Employs simple string manipulation (e.g., sed, cut) but lacks robust validation or escaping of input data used in shell commands.

vercel-cli-with-tokens