triaging-live-sql-activity
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes live SQL query text from the cluster, which is an untrusted data source that could contain embedded instructions. \n
- Ingestion points: Output from cluster activity views (SHOW CLUSTER STATEMENTS, SHOW CLUSTER SESSIONS, crdb_internal.cluster_transactions) defined in SKILL.md and references/sql-queries.md. \n
- Boundary markers: The instructions do not define specific delimiters for isolating database output from the agent's control context. \n
- Capability inventory: The agent has the capability to generate and execute administrative SQL commands such as CANCEL QUERY. \n
- Sanitization: SQL queries use the substring() function to truncate query previews, which limits the footprint of potential injection payloads. \n- [SAFE]: The skill uses official CockroachDB SQL interfaces for all administrative triage tasks. The author and content are consistent with legitimate vendor tools, and documentation provides best-practice guidance on privilege management and privacy redaction.
Audit Metadata