The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external database schemas (table names, column comments, view definitions) which acts as an untrusted input source. If a database contains malicious strings in comments or metadata, it could influence the agent's behavior.
Ingestion points: Steps 1-7 in SKILL.md query information_schema and pg_catalog for metadata.
Boundary markers: None identified; instructions do not explicitly tell the agent to ignore embedded instructions in the database metadata.
Capability inventory: The skill uses Bash(psql:*), Bash(mysql:*), and Write tools to interact with the database and save documentation.
Sanitization: No explicit sanitization or escaping of database content (like comments) is mentioned before inclusion in the final documentation.
[COMMAND_EXECUTION]: The skill utilizes shell-based database clients (psql, mysql, mongosh) to perform introspection via the Bash tool. While restricted in scope, these commands interact with live external environments.

database-documentation-gen