The Agent Skills Directory

[SAFE]: The skill instructions and reference materials are focused entirely on database performance tuning and do not contain any malicious or suspicious patterns. All suggested SQL and shell commands are standard tools for query analysis and benchmarking.
[DATA_EXFILTRATION]: There are no indicators of unauthorized data access or network exfiltration. The skill is designed to perform diagnostic analysis within the database environment and report findings back to the user or other agents.
[REMOTE_CODE_EXECUTION]: The skill does not perform any remote script downloads or executions. It references well-known and standard database utilities like pgbench and sysbench for benchmarking purposes.
[COMMAND_EXECUTION]: Command execution is limited to standard database administrative tasks (e.g., CREATE INDEX CONCURRENTLY, ALTER SYSTEM) and monitoring queries, all of which are appropriate for a database tuning specialist role.
[PROMPT_INJECTION]: No evidence of jailbreak attempts, safety guideline overrides, or role-play injection patterns were found in the skill definition or metadata.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests external data (database query plans and schema information), the risk is minimal as the output is primarily structured recommendations for a human or another agent to review.
Ingestion points: SKILL.md (Analyze phase processes query plans, slow-query samples, and workload context).
Boundary markers: Use of Markdown code blocks for structured data.
Capability inventory: SQL execution (EXPLAIN ANALYZE), configuration modification (ALTER SYSTEM), and benchmark tool invocation (pgbench, sysbench).
Sanitization: Not explicitly defined in documentation, but the skill follows a consultative workflow where recommendations are proposed rather than executed blindly.

Tuner