The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides the ability to read and decrypt Vercel environment variables using the env get and env shared get commands. While this is a core project management feature, it grants the agent access to potentially sensitive application secrets and credentials stored in the Vercel platform.
[COMMAND_EXECUTION]: The skill executes actions via a specialized agent-vercel CLI tool. This includes an "escape hatch" command, agent-vercel api call, which allows the agent to perform arbitrary REST API requests against the Vercel platform. Although non-GET requests require a --yes flag, this provides a broad capability surface for interacting with project infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it retrieves and processes untrusted data from Vercel build and runtime logs.
Ingestion points: Untrusted data enters the agent's context through deployment logs and deployment runtime-logs (documented in SKILL.md and references/commands.md).
Boundary markers: The instructions do not specify the use of delimiters or specific "ignore embedded instructions" warnings when processing log content.
Capability inventory: The agent has the capability to perform state-changing operations such as deployment promote, deployment rollback, env set, and arbitrary API calls (documented in SKILL.md).
Sanitization: There is no mention of sanitization, filtering, or escaping for external content retrieved from logs before it is processed by the agent.

agent-vercel