The Agent Skills Directory

[SAFE]: The skill implements a robust credential management policy. Vercel access tokens are stored in the macOS Keychain and the instructions explicitly forbid the agent from retrieving these tokens or asking the user to provide them in the chat interface.
[SAFE]: Destructive or state-changing operations, such as deployment promotions, rollbacks, and environment variable modifications, are protected by a mandatory --yes flag. The agent is instructed to describe the proposed action and obtain user confirmation before executing with the gate flag.
[SAFE]: The skill provides clear privacy guidance, ensuring that sensitive platform data like access tokens are never printed. Even in 'full' output modes, specific fields like delivery URLs in log drains (which may contain destination tokens) are omitted or handled with care.
[SAFE]: While the skill can read application environment variables using the --decrypt flag, this is a core intended functionality for a deployment triage tool and is constrained by the credential's scope and safety instructions.
[SAFE]: The tool contains an attack surface for indirect prompt injection because it reads build and runtime logs from external Vercel deployments. However, this risk is effectively mitigated by the mandatory human-in-the-loop requirement (via the --yes safety gate) for any subsequent actions the agent might take based on that data.

agent-vercel