dunning-cohort-builder

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external API, creating a potential surface for indirect prompt injection attacks.
  • Ingestion points: Fetches subscription records from ZohoBilling_get_subscriptions_dunning_report in SKILL.md (Step 1).
  • Boundary markers: Absent; data fields like gateway_error_message are processed and passed downstream without specific delimiters to distinguish between data and instructions.
  • Capability inventory: The skill aggregates and summarizes data, then passes a structured JSON cohort to the dunning-action-planner tool.
  • Sanitization: Includes basic lowercase and trim operations on gateway codes and messages in Step 2, but does not implement advanced filtering for natural language instructions.
  • [SAFE]: The tool calls are limited to the vendor's own infrastructure (Zoho Billing), and the logic consists of standard data normalization and arithmetic scoring. No obfuscation, persistence, or privilege escalation patterns were found.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 05:18 AM
Security Audit — agent-trust-hub — dunning-cohort-builder