The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data from MySQL databases, creating a potential surface for indirect prompt injection where database content could influence agent behavior.
Ingestion points: Database query results retrieved via pool.execute() and pool.query() in examples/core.md and examples/streaming.md.
Boundary markers: While SQL parameterization is strictly enforced to protect the database, the skill does not specify using boundary markers (e.g., XML tags or clear delimiters) when the agent presents retrieved data to the user or passes it to other tools.
Capability inventory: Includes database read/write access and local file system write operations (demonstrated in examples/streaming.md).
Sanitization: The skill emphasizes input sanitization via ? placeholders to prevent SQL injection, though it does not detail sanitization or validation of data retrieved from the database before it is processed by the LLM.
[DATA_EXFILTRATION]: The skill demonstrates standard security practices for data handling, such as using environment variables (DATABASE_URL) for connection strings rather than hardcoding credentials. It also utilizes node:fs for legitimate operations like reading SSL/TLS certificates and writing data streams to local files.
[EXTERNAL_DOWNLOADS]: The skill utilizes the well-known and trusted mysql2 Node.js package for database connectivity.

api-database-mysql