The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to check for the presence of the Vercel CLI (command -v vercel) and to execute deployment tasks (vercel deploy). It also runs a local script, scripts/deploy.sh, to handle fallback deployment logic when CLI authentication is not available.
[DATA_EXFILTRATION]: The scripts/deploy.sh script packages the project into a tarball and uploads it to an external deployment API (https://codex-deploy-skills.vercel.sh/api/deploy). The script implements privacy protection by explicitly excluding sensitive directories and files, such as .git, node_modules, and environment configuration files (.env, .env.*), from the deployment package.
[PRIVILEGE_ESCALATION]: The skill documentation in SKILL.md includes guidance for the agent to request escalated network permissions (sandbox_permissions=require_escalated) if the deployment process is restricted by sandboxing rules in the execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it reads and parses project-level configuration files to automate deployment settings.
Ingestion points: The scripts/deploy.sh script reads the project's package.json file (line 18) to determine the appropriate framework for deployment.
Boundary markers: No specific delimiters or safety instructions are used to separate the project data from the execution environment's context.
Capability inventory: The skill has capabilities for network communication (curl), filesystem packaging (tar), and command execution (vercel CLI).
Sanitization: Framework detection is performed via simple pattern matching on dependency names within the file content, without advanced sanitization of the input data.

vercel-deploy