The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents the installation of the postgresql-client utility using official system package managers, including Homebrew (brew), apt, yum, and apk (SKILL.md). These references target well-known, trusted operating system repositories.
[SAFE]: Recommends the use of the ~/.pgpass file for managing database credentials, which is a standard security best practice for PostgreSQL to prevent passwords from being exposed in process listings or command history (SKILL.md).
[PROMPT_INJECTION]: The skill documents psql features that allow the agent to process database output and execute shell commands, which creates a potential surface for indirect prompt injection. This risk is effectively mitigated by the skill's extensive safety instructions.
Ingestion points: The agent may ingest untrusted data from database query results or schema metadata retrieved via meta-commands like \d (references/meta-commands-inspection.md).
Boundary markers: While no specific data delimiters are defined for result parsing, the skill provides a "Destructive Operations Checklist" and explicit safety warnings for dangerous meta-commands (SKILL.md).
Capability inventory: The documented toolset includes powerful capabilities such as arbitrary shell execution (\!), piping output to shell commands (\o |cmd, \w |cmd), and executing query results as SQL (\gexec) as detailed in references/meta-commands-core.md.
Sanitization: The skill explicitly instructs the agent on sanitization best practices, such as using \bind for truly parameterized queries and proper variable quoting (:'varname', :"varname") to prevent SQL injection (references/tips-advanced.md).

postgresql-cli