aviation-cleanup

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is designed to generate and execute SQL DROP and ALTER statements to remove Snowflake objects like databases, schemas, tasks, and integrations. This is the primary function of the skill and is accompanied by instructions for user review and dry-run execution.\n- [PROMPT_INJECTION]: The skill processes JSON metadata from Snowflake object comments to identify objects for deletion, creating a surface for indirect prompt injection.\n
  • Ingestion points: Object comments retrieved from INFORMATION_SCHEMA and SHOW commands (SKILL.md, discovery-queries.sql).\n
  • Boundary markers: None; the agent is instructed to parse the origin tag from the comment field.\n
  • Capability inventory: Execution of destructive SQL commands and local file system writes for logging (SKILL.md, drop-order.sql).\n
  • Sanitization: Discovery queries use TRY_PARSE_JSON for validation, and the workflow requires explicit user confirmation before any modifications occur.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 10:03 PM
Security Audit — agent-trust-hub — aviation-cleanup