tablet
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes database metadata, such as table names, partition names, and tablet identifiers, and incorporates these values into administrative command templates to diagnose and resolve performance issues.
- Ingestion points: Metadata is ingested from
information_schema.be_tablets,information_schema.tables, and the output ofSHOW TABLETandSHOW PROCcommands as described inSKILL.mdandreferences/03-tablet-schema-queries.md. - Boundary markers: The instructions do not define specific delimiters or directives to treat database metadata as untrusted, creating a surface where malicious instructions embedded in metadata names could be processed by the agent.
- Capability inventory: The skill includes instructions for high-privilege operations, including system configuration updates (
UPDATE information_schema.be_configs), replica status modification (ADMIN SET REPLICA STATUS), and backend API triggers viacurlPOST requests. - Sanitization: There is no explicit sanitization or validation logic provided for the metadata identifiers before they are interpolated into administrative command templates.
- [COMMAND_EXECUTION]: The skill provides numerous templates for shell commands using
curlto interact with the StarRocks Backend (BE) HTTP API. These commands are intended for administrative diagnostics and manual compaction triggers, targeting internal backend addresses.
Audit Metadata