The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external contract artifacts such as OpenAPI or JSON Schema files, which are untrusted data sources. It proactively mitigates injection risks by instructing the agent to treat all embedded content as data rather than instructions, and to sanitize $ref targets to prevent path traversal or unauthorized remote references.
[COMMAND_EXECUTION]: The skill mentions executing standard repository scripts like npm run generate:api-types. It explicitly advises running these generators with least privilege, recommending the removal of network and secret access and limiting write permissions to specific output directories.
[SAFE]: The workflow emphasizes security best practices, including using pinned dependencies, verifying serialized data at runtime boundaries rather than relying solely on compile-time types, and maintaining a single source of truth for interfaces to prevent drift-related vulnerabilities.

contract-first