The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied table and schema names for export jobs, creating an attack surface for identifier-based injection.
Ingestion points: table and schema fields defined in the Payload JSON section of SKILL.md.
Boundary markers: The instructions explicitly require the agent to validate these identifiers against a strict regular expression (^[A-Za-z_][A-Za-z0-9_]*$).
Capability inventory: The skill uses POST requests to the worker API (geobase.app) and executes vendor CLI commands (geobase-cli).
Sanitization: Mandatory regex validation is instructed to prevent non-alphanumeric characters or injection sequences from entering the payload.
[SAFE]: The skill demonstrates high security awareness regarding credential management. It explicitly instructs the agent not to attempt to obtain the GEOBASE_SERVICE_ROLE_KEY and requires the user to manage secrets in a gitignored file, ensuring a human-in-the-loop security model for high-privilege access keys.

geobase-worker-archive-to-pmtiles