israeli-customer-support-automator
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and categorize untrusted Hebrew text from customer support tickets, creating a surface for indirect prompt injection. A malicious ticket could attempt to influence agent behavior or categorization logic. Findings: Untrusted data is ingested via ticket text in SKILL.md and scripts/ticket-classifier.py; there are no explicit boundary markers to delimit untrusted input; the agent has capabilities for shell command execution and escalation; no input sanitization or filtering is implemented before processing.
- [EXTERNAL_DOWNLOADS]: The skill references external informational resources from several trusted organizations and well-known services, including Israeli government portals (gov.il), legal databases (nevo.co.il, kolzchut.org.il), and official platform documentation (developers.facebook.com). These references are consistent with the skill's primary purpose.
- [COMMAND_EXECUTION]: The skill uses the Bash tool to execute a local analysis script (scripts/ticket-classifier.py) to perform keyword-based ticket classification. The script uses only standard library components and performs local analysis on ticket content.
Audit Metadata