The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides instructions for building and executing SQL queries using user-provided inputs, which constitutes a surface for indirect prompt injection or SQL injection.
Ingestion points: Examples in the 'Security Patterns', 'AST Manipulation Patterns', and 'Application Service Pattern' sections of SKILL.md demonstrate the processing of user-supplied table names, tenant IDs, and query filters.
Boundary markers: The documentation advocates for the use of AST-based parsing and validation via sqlglot to ensure input is treated as a safe identifier rather than executable code.
Capability inventory: The database drivers and adapters described in the skill allow for the execution of SQL statements across multiple backend systems (e.g., Postgres, MySQL, SQLite).
Sanitization: The skill includes specific examples for sanitizing table names and detecting dangerous SQL patterns (e.g., DROP, TRUNCATE) using sqlglot before query execution.
[EXTERNAL_DOWNLOADS]: The skill references several external websites and repositories, including official documentation for sqlspec, sqlglot, and mypyc. It also links to styleguides hosted on the author's GitHub repository (github.com/cofin/flow). These references are standard for technical documentation and align with the stated purpose of the library.

sqlspec