shop-push-broadcast

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill implements a strictly ordered workflow with a mandatory 'Explicit confirmation' step before performing the sensitive action of sending a push broadcast to all shop users.
  • [SAFE]: Guardrails are in place to prevent automatic sends, empty messages, or placeholder content (e.g., 'test', 'lorem').
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data from product and CMS catalogs, creating a potential surface for indirect prompt injection if external metadata contains malicious instructions.
  • Ingestion points: Tool outputs from shop_list_products, cms_list_cms_sections, and cms_list_sections (SKILL.md).
  • Boundary markers: No explicit XML/markdown delimiters are used to wrap ingested data, but the skill requires a specific structured output format.
  • Capability inventory: Sending push notifications to the entire registered user base via shop_create_push_broadcast (SKILL.md).
  • Sanitization: The skill includes mandatory spellchecking, character limit enforcement (<= 255 chars), and a hard requirement for explicit user confirmation ('Confirm send?') before the final tool call, which effectively mitigates the risk of unintended broadcasts triggered by data injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 02:57 PM
Security Audit — agent-trust-hub — shop-push-broadcast