The Agent Skills Directory

[SAFE]: The skill follows a 'local-first' design, instructing the agent to perform GraphQL architecture work locally and use external SandBase tools only for documentation or explicitly requested model inference.
[SAFE]: The workflow includes mandatory human-in-the-loop checkpoints, requiring the agent to present tool arguments and pricing estimates before obtaining explicit user confirmation for paid or external calls.
[SAFE]: The instructions actively promote security best practices for GraphQL implementation, including resolver-level authorization, query complexity bounding, and N+1 access prevention, which are standard defensive measures for the task domain.
[SAFE]: The identified tools (sandbase_discover, sandbase_inspect, sandbase_run, etc.) are vendor-specific resources associated with the skill author 'sandbaseai' and are used in a controlled, gated manner.

graphql-architect