The Agent Skills Directory

[SAFE]: The skill performs legitimate database security auditing. It includes detailed safety rules for production environments, requiring all verification steps to be performed within transactions that terminate with ROLLBACK or RAISE EXCEPTION to prevent any accidental state changes.
[DATA_EXPOSURE_SAFE]: The skill incorporates specific guidelines to prevent data exposure during the audit process, explicitly instructing the agent to report the length or hash of sensitive bearer tokens rather than the tokens themselves.
[INDIRECT_PROMPT_INJECTION_SAFE]: While the skill ingests external data (database catalog metadata and application code via ripgrep), the risk is mitigated by the structured, procedure-driven nature of the audit workflow which defines clear schemas and expected outputs, reducing the likelihood of the agent following instructions embedded in table comments or code strings.
[COMMAND_EXECUTION_SAFE]: The skill suggests the use of standard search tools like rg (ripgrep) for identifying code consumers of database tokens. These are used appropriately for static analysis of the local codebase as part of the primary auditing purpose.

postgres-privilege-audit