The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill includes instructions for modifying system-level kernel parameters and configuration files which require root/administrative privileges.
Evidence in references/config-tuning.md: echo 9000 > /proc/sys/vm/nr_hugepages, echo "vm.nr_hugepages = 9000" >> /etc/sysctl.conf, and sysctl -p.
Evidence in references/operations.md: Commands to move the database data directory (mv /var/lib/postgresql/14/main /var/lib/postgresql/14/main.bak), which typically requires high-level system access.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform database operations, including backups and restores.
Evidence in SKILL.md and references/operations.md: Usage of pg_dump, pg_restore, and pg_basebackup.
[INDIRECT_PROMPT_INJECTION]: The skill describes workflows for analyzing database schemas, query plans, and monitoring metrics, which involves ingesting potentially untrusted data that could influence the agent's behavior.
Ingestion points: SQL query results, schema definitions, and system metrics extracted via psql or monitoring extensions.
Boundary markers: The instructions do not define boundary markers or explicit warnings to the agent regarding embedded instructions in the data it processes.
Capability inventory: The skill has access to Bash, Read, and Write tools, enabling it to execute commands or modify files based on its analysis.
Sanitization: There is no evidence of sanitization or validation routines for data retrieved from the database context.

postgres-ops