The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script (scripts/pgdiag) that executes the psql binary to collect database statistics and configuration. The command is constructed as a list of arguments, which is a secure practice to prevent shell injection. Additionally, the script hardcodes a read-only session setting (default_transaction_read_only = on) to prevent accidental data modification.
[DATA_EXPOSURE]: The diagnostic tool collects PostgreSQL metadata, configuration parameters, and performance metrics. The skill's documentation explicitly instructs the agent and user not to share sensitive information such as passwords or full configuration files within the chat environment.
[REMOTE_CODE_EXECUTION]: The pgdiag script allows specifying the path to the psql binary via a command-line flag. While this allows execution of a local file, it is standard for database administrative tools and does not involve downloading or executing code from external, untrusted network sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from database catalogs (table names, index names, and statistics). A malicious actor with database access could attempt to inject instructions into these metadata fields. However, the skill outputs this data in a structured JSON format and requires human confirmation for any subsequent database mutations, effectively mitigating the risk of automated exploitation.

postgres