The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation describes the use of CLI-based actions like pnpm action db-query, db-exec, and db-patch. These tools allow the agent to execute raw SQL queries, perform updates, and conduct surgical search-and-replace operations on large text columns in the application database.
[PROMPT_INJECTION]: The skill establishes a data ingestion pipeline where the agent is encouraged to read from SQL tables such as application_state, settings, and domain-specific tables (e.g., tasks, emails). This represents a surface for indirect prompt injection if the stored data originates from untrusted sources.
Ingestion points: Tool outputs from db-query and framework functions like readAppState(), getSetting(), and useActionQuery which pull data from persistent and ephemeral stores into the agent's context.
Boundary markers: The instructions do not define specific delimiters or instructions to treat database content as untrusted when processing it via the agent.
Capability inventory: The agent possesses powerful tools for database modification (db-exec, db-patch), file management (uploadFile, putPrivateBlob), and reference is made to the agent's ability to modify source code via the self-modifying-code skill.
Sanitization: While the skill emphasizes using Drizzle ORM's query builder to prevent SQL injection during database writes, it lacks specific guidance on sanitizing or validating retrieved data before it is interpreted by the agent's language model.

storing-data