The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/loader.sh script downloads the aurora-dsql-loader binary from the official aws-samples GitHub repository. This process is secured with HTTPS enforcement, file size checks, and binary header validation to prevent the execution of corrupt or malicious files.
[COMMAND_EXECUTION]: Multiple bash scripts in the scripts/ directory (e.g., create-cluster.sh, psql-connect.sh, loader.sh) execute AWS CLI and psql commands. These scripts use array-based command construction and jq for JSON processing to safely handle arguments and prevent shell injection vulnerabilities.
[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection via its database interaction tools (readonly_query and transact). As documented in mcp-tools.md, these tools do not support parameterized queries, meaning SQL strings constructed from untrusted external data could be exploited for SQL injection. The skill includes mandatory evidence for this surface: (1) Ingestion points: SQL queries passed to MCP tools in mcp-tools.md; (2) Boundary markers: None present for the queries themselves; (3) Capability inventory: Full DML and DDL capabilities via the transact tool; (4) Sanitization: Not performed by the tools; requires manual implementation by the agent using regex and allowlist patterns.

dsql