The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with the Vercel ecosystem by executing the vercel CLI binary. It enables complex workflows including deployment management and project configuration through delegated shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted external data, specifically build and runtime logs via observability.read and deployment file contents via deployment_files.read.
Ingestion points:build/runtime logs, deployment file contents, and sandbox sessions.
Boundary markers: The skill relies on connector-level permissions and user confirmation rather than specific prompt delimiters for data isolation.
Capability inventory: The skill possesses extensive capabilities including CLI execution (cli.execute), project creation (projects.create), and secret management (secrets.read).
Sanitization: Security is managed through a granular permission model where high-risk read and write operations default to a user-review (ask) state.
[DATA_EXFILTRATION]: The skill has the capability to access highly sensitive information such as billing records, environment variables, and decrypted secrets. This risk is mitigated by the manifest configuration which requires explicit user consent (default: ask) before accessing these specific data scopes.

vercel