The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes instructions for administrative tasks such as restarting the ClickHouse service using sudo systemctl restart or docker restart, and inspecting kernel/service logs via dmesg and journalctl.
[COMMAND_EXECUTION]: Employs destructive database operations, including ALTER TABLE ... DROP PARTITION for managing disk space and KILL QUERY for terminating long-running processes.
[DATA_EXFILTRATION]: Instructs the agent to write database query logs and metrics to local files at /tmp/incident-queries.json and /tmp/incident-metrics.tsv, which may expose sensitive query information and performance metadata to other users or processes on the host.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by processing untrusted content from database system tables.
Ingestion points: Reads from system.query_log and system.processes in SKILL.md to collect evidence and triage issues.
Boundary markers: Does not utilize markers to isolate database output from agent instructions.
Capability inventory: Possesses high-privilege access to shell environments, system service management via sudo, and administrative database controls.
Sanitization: Lacks mechanisms to sanitize or validate database log entries before the agent processes them.
[EXTERNAL_DOWNLOADS]: Retrieves service health information from the official ClickHouse Cloud status domain.

clickhouse-incident-runbook