The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from a PostgreSQL database through the queryPgDatabase tool. The instructions do not provide explicit guidance for the agent to treat these tool outputs as untrusted data or to utilize boundary markers when processing database records. This creates an attack surface where malicious instructions stored in the database could influence agent behavior, especially given the skill's extensive capabilities for database and environment management.
Ingestion points: SQL query results and schema information retrieved via queryPgDatabase (referenced in SKILL.md and app-workflow.md).
Boundary markers: The instructions do not define delimiters or specific warnings for the agent when processing data returned from the database.
Capability inventory: The skill provides access to powerful tools including managePgDatabase (SQL execution), manageAppAuth (credential management), and manageEnv (environment configuration).
Sanitization: No explicit logic or filtering instructions are provided to sanitize external content retrieved from database tables before interpolation into prompts.
[SAFE]: All external URLs and resources mentioned, such as tcloudbasegateway.com and cloudbase.net, are official infrastructure belonging to the vendor tencentcloudbase and do not represent a security risk.
[SAFE]: The skill promotes security best practices, specifically warning against the exposure of service_role API keys in frontend code and providing detailed templates for Row-Level Security (RLS) and table-level GRANTs.

postgresql-development-cloudbase