The Agent Skills Directory

[COMMAND_EXECUTION]: Privilege Escalation via Database Permissions. The instructions in SKILL.md direct users to grant SUPERUSER privileges or add accounts to the pg_read_all_stats group. These permissions are required to access system-level logs in the hologres.hg_query_log table but grant extensive control over the database instance.\n- [COMMAND_EXECUTION]: The skill relies on the hologres-cli tool to execute SQL queries and modify database-level GUC parameters (e.g., hologres sql run, hologres guc set). This provides a path for the agent to execute commands on the underlying database infrastructure.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill analyzes query logs that contain arbitrary SQL strings and error messages provided by database users.\n
Ingestion points: The system table hologres.hg_query_log (referenced in SKILL.md and references/diagnostic-queries.md) serves as the entry point for untrusted data.\n
Boundary markers: No specific delimiters or instructions are provided to the agent to treat log content as untrusted data.\n
Capability inventory: The agent has the capability to execute database commands and modify configurations via hologres-cli.\n
Sanitization: There is no evidence of sanitization or filtering applied to log data before it is presented to or processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the hologres-cli Python package via pip. This is a vendor-provided tool for managing Alibaba Cloud Hologres resources.\n- [DATA_EXFILTRATION]: Workflows in references/log-export.md describe how to export sensitive slow query logs to external storage services like MaxCompute or OSS. This is a standard administrative task but involves moving sensitive database activity logs.

hologres-slow-query-analysis