The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/deploy.sh script packages the project directory and uploads it to an external endpoint: https://codex-deploy-skills.vercel.sh/api/deploy. While the script excludes .env and .git directories, the transmission of the full source code to a non-official Vercel API endpoint represents a data exposure risk.
[PRIVILEGE_ESCALATION]: Instructions in SKILL.md direct the agent to utilize sandbox_permissions=require_escalated if deployment fails due to networking restrictions, effectively requesting a bypass of standard sandbox constraints.
[COMMAND_EXECUTION]: The skill executes multiple shell commands via scripts/deploy.sh for project packaging (tar), framework detection (grep), and deployment (curl), and interacts with the vercel CLI if available.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: scripts/deploy.sh reads and processes the project's package.json file to detect framework types.
Boundary markers: None identified for file content processing.
Capability inventory: The script has capabilities to read local files, create archives, and perform network requests.
Sanitization: No specific sanitization is performed on the data read from package.json before it is used in logic checks.

vercel-deploy