The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a pattern for the agent to ingest and reason over data from a large number of tenant databases (~180). This creates an indirect prompt injection surface where instructions embedded in processed database records (such as case notes, vehicle reports, or LPR data) could influence the agent's behavior.
Ingestion points: The agent utilizes execute_sql and search_objects tools to fetch data from Insightt (MySQL) and Foundd (PostgreSQL) databases.
Boundary markers: Instructions do not include specific delimiters or guidelines to distinguish untrusted database content from the agent's core instructions.
Capability inventory: The agent is granted capabilities to perform SQL queries and explore database schemas across hundreds of tenants.
Sanitization: No sanitization or validation protocols are defined for the data retrieved from the databases.
[DATA_EXPOSURE_AND_EXFILTRATION]: The reference documentation in insightt.md describes the schema for routing tables (platform_companies) that store per-tenant database credentials, including usernames and passwords (db_username, db_user_password). While this information is intended to facilitate system routing, it identifies a high-value target for potential credential harvesting if the agent is directed to dump these tables.

multi-tenant-db