The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from the Supabase environment, such as table names from database statistics and bucket/object names from storage, to perform its analysis.
Ingestion points: Audit queries in references/audit-queries.md and storage listing logic in references/examples.md pull external identifiers into the agent context.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat identifiers (like table names) as data rather than instructions.
Capability inventory: The skill has access to Write, Edit, and Bash(supabase:*) tools, which could potentially be abused if the agent is misled by malicious metadata.
Sanitization: No validation or sanitization of database identifiers is performed before they are used in generated optimization commands like VACUUM or DROP INDEX.

supabase-cost-tuning