The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface detected. The skill instructs the agent to process potentially untrusted data from the database server without providing security boundaries or sanitization instructions.
Ingestion points: Untrusted data enters the agent context via database system views and logs, specifically pg_stat_statements, pg_stat_io, EXPLAIN output, and slow query logs as noted in SKILL.md and references/query-plan-analysis.md.
Boundary markers: The skill lacks recommendations for using delimiters (e.g., XML tags or markdown blocks) to isolate untrusted data from instruction prompts.
Capability inventory: The skill suggests performing sensitive database operations such as EXPLAIN ANALYZE (which executes code), CREATE INDEX, and ALTER TABLE through the Task agent.
Sanitization: No guidance is provided to sanitize or validate the query text or plan outputs before the agent processes them.

postgres-tuning