The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data when importing resources from a RudderStack workspace. A vulnerability surface exists where user-generated content, such as event descriptions, could contain malicious instructions.
Ingestion points: Data enters the environment via the rudder-cli import workspace command, which populates the imported/ directory with YAML files from the RudderStack cloud.
Boundary markers: No technical delimiters are used for the imported content, but the skill includes a specific 'Handling External Content' section to guide the agent and user.
Capability inventory: The skill utilizes rudder-cli apply to modify cloud workspace state and has access to Write and Edit tools for local file management.
Sanitization: The skill provides explicit mitigation steps including manual review of imported YAML, verification of structure against known schemas, and sanitization of user-generated descriptions before committing to version control.
[COMMAND_EXECUTION]: The skill uses the vendor's official rudder-cli tool for all operations. Security is maintained by requiring the use of environment variables for authentication rather than interactive logins, and by mandating a dry-run review by the user before any changes are applied using the --confirm=false flag.

rudder-import-and-evolve