The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include a script to run bun install --frozen-lockfile to restore dependencies. It also provides a connections.ts utility intended to be executed via the command line for managing PostgreSQL connection profiles.
[DYNAMIC_EXECUTION]: The interactive quickstart guide in SKILL.md uses dynamic import() calls to load the skill's management script and the pg (node-postgres) library at runtime. It also uses Bun.resolveSync to determine package paths dynamically.
[INDIRECT_PROMPT_INJECTION]: The skill exposes powerful database interaction functions (pgQuery and pgTransaction) to the AI agent. If the agent is instructed to process untrusted external data and subsequently use that data to construct SQL queries, there is a risk of SQL injection or unauthorized data access. The provided functions support parameterization (values argument), which is a key mitigation, but the overall capability remains high-risk if misused.
Ingestion points: The agent context, where instructions or external data are processed before being used in database queries.
Boundary markers: Not explicitly defined for the SQL queries; the agent is responsible for correctly utilizing the parameterization features of the provided functions.
Capability inventory: Execution of arbitrary SQL queries (pgQuery), transaction management (pgTransaction), and management of connection credentials.
Sanitization: The skill provides a parameterized query interface (pgQuery(text, values)), which facilitates safe SQL practices if correctly utilized by the agent.
[DATA_EXFILTRATION]: The skill is explicitly designed to transmit data to and from external PostgreSQL servers. It handles sensitive credentials by storing them in the .connections/ directory. The skill adopts defensive practices by setting file permissions to 0o600 and ensuring passwords are redacted from standard log output, following standard secret management patterns.

bayma-platform-postgres