The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from database query results and infrastructure metrics which serves as an attack surface for indirect prompt injection.
Ingestion points: The agent reads data from SQL views (e.g., pg_stat_statements, pg_stat_activity, ASH) and external metrics sources including Prometheus and the YugabyteDB Anywhere (YBA) API.
Boundary markers: The instructions include strong directives for the agent to remain "read-only by default" and explicitly forbid executing any state-mutating commands (DDL, restarts, session termination) without prior human approval.
Capability inventory: The skill utilizes kubectl exec, ysqlsh, and HTTP POST requests to the YBA metrics proxy to fetch diagnostic information.
Sanitization: The skill focuses on generating a ranked report for human review rather than autonomous execution of recommendations.
[COMMAND_EXECUTION]: The skill leverages standard command-line utilities to interact with database clusters and orchestration platforms.
Evidence: Usage of kubectl exec to enter tserver pods and ysqlsh/psql for database connectivity.
Context: These are intended capabilities essential for the primary purpose of database performance triage and are used in a diagnostic, read-only capacity.

yb-performance-assessment