The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze metadata and sample data from external databases, which represents a potential attack surface if the database content is maliciously crafted.\n
Ingestion points: Results from exploration queries targeting information_schema, pg_catalog, and system columns, as well as sample rows from user databases, as described in references/exploration.md.\n
Boundary markers: The skill mandates a structured evidence-based reasoning framework (OBSERVED, INFERRED, CONFIRMED) in SKILL.md and references/data-contract.md to ensure that raw metadata does not bypass review.\n
Capability inventory: The skill can generate and, with explicit user permission, execute SQL DDL commands across multiple database dialects (MySQL, PostgreSQL, ClickHouse, Doris, etc.) as documented in references/partitioning.md.\n
Sanitization: Instructions in references/exploration.md require query budget limits and data sampling restrictions to prevent excessive data exposure.\n- [COMMAND_EXECUTION]: The skill generates complex SQL DDL scripts and supports their execution against connected databases upon explicit user request.\n
Evidence: The skill defines a complete workflow for generating Dialect-specific DDL for schema creation, partitioning, and migration in SKILL.md and references/partitioning.md.\n- [EXTERNAL_DOWNLOADS]: The skill references documentation and open standards from public GitHub repositories for its data contract and SQL styling guidelines.\n
Evidence: references/external-standards.md provides links to github.com/bitol-io/open-data-contract-standard and github.com/TFMV/datacontract-specification which are used as reference architectures for the skill's deliverables.

finance-ddl-design