The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The dm8_query.py script allows execution of arbitrary SQL statements. While this is the intended purpose, the tool permits destructive operations (e.g., DELETE, DROP, UPDATE) and does not use parameterized queries. This poses a risk if the agent is manipulated by prompt injection to execute malicious commands.
[REMOTE_CODE_EXECUTION] (MEDIUM): The skill relies on a binary Java archive (DmJdbcDriver18.jar) to function. This file is mentioned as 'built-in' in SKILL.md but is missing from the provided source files, making it an unverifiable binary dependency. Loading external binary drivers into a JVM via JPype1 and jaydebeapi is a common vector for code execution.
[CREDENTIALS_UNSAFE] (LOW): Database passwords are required as command-line arguments (--password) for all scripts. This is a poor security practice as it can expose credentials in process lists, shell history, and system logs.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill provides a significant surface for indirect prompt injection by reading untrusted data from database tables.
Ingestion points: Results from dm8_query.py, dm8_tables.py, and dm8_schema.py are returned directly to the agent.
Boundary markers: None. The output is printed as a raw JSON string without delimiters.
Capability inventory: Full SQL execution capabilities (read/write/delete) and network access to database hosts.
Sanitization: No sanitization or escaping is performed on the data retrieved from the database before it is provided to the agent context.

dm8-tools