The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to its interaction with external Vercel CLI outputs. \n
Ingestion points: The skill reads data from the Vercel platform using commands such as vercel logs, vercel ls, and vercel inspect (SKILL.md). \n
Boundary markers: No specific delimiters are defined in the instructions to distinguish between trusted CLI output and potential malicious instructions embedded in logs or project names. \n
Capability inventory: The skill possesses destructive capabilities, including deleting deployments (vercel rm) and projects (vercel projects rm), and modifying environment variables (vercel env add/rm) (SKILL.md). \n
Sanitization: No explicit sanitization or validation logic for CLI output is mentioned. \n
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill references the installation and updating of the Vercel CLI. \n
Evidence: Prerequisite step npm i -g vercel and the vercel upgrade command (SKILL.md). \n
Context: These operations target Vercel, which is a well-known technology service, and are considered safe standard operations for this skill's use case. \n
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages sensitive configuration data including environment variables and secrets. \n
Evidence: Use of commands like vercel env add, vercel secrets add, and reference to --token $VERCEL_TOKEN (SKILL.md). \n
Context: This is the primary intended purpose of the skill for managing cloud deployments and does not involve unauthorized exfiltration.

vercelops