The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes SQL statements from untrusted external sources, including database dumps, ORM migrations, and user-provided DDL. This represents a surface for indirect prompt injection where malicious instructions could be embedded in SQL comments or data structures to influence agent behavior.
Ingestion points: SQL input, externally-sourced SQL files (pg_dump, ORM migrations) referenced in SKILL.md.
Boundary markers: The skill mandates the use of dsql_lint for validation and safe_query.build() for query construction.
Capability inventory: Direct SQL execution via MCP tools (readonly_query, transact), local bash script execution (scripts/), and shell command execution via psql fallback.
Sanitization: Implements a SQL builder (safe_query.py) that uses allow-lists and regex for values and dedicated identifier quoting for table/column names.
[COMMAND_EXECUTION]: The skill provides bash scripts in the scripts/ directory for cluster management (creation, deletion, listing) and bulk data loading. It also defines a fallback procedure that executes shell commands using the aws CLI to generate IAM authentication tokens and the psql client for database connectivity.
[EXTERNAL_DOWNLOADS]: The skill references documentation and code samples from official AWS domains (docs.aws.amazon.com) and the awslabs GitHub organization. These references target trusted vendor resources and are documented neutrally.

amazon aurora dsql