israeli-address-autocomplete
Warn
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill contains a hardcoded API subscription key (
5ccb5b137e7444d885be752eda7f767a) for an undocumented Israel Post API endpoint. While intended for address lookups, hardcoding credentials in instructions is a violation of secret management practices. - [EXTERNAL_DOWNLOADS]: The instructions direct the agent to fetch remote JavaScript bundles from a bot-protected website (
doar.israelpost.co.il) and perform pattern matching to extract active subscription keys. This technique is designed to bypass authentication controls and intended rate-limiting or bot protections for the underlying API. - [METADATA_POISONING]: The skill includes files like
optimization-log.jsonandevidence.jsonthat use technical jargon (e.g., "Phase 5.7 gate", "Expert Review", "Radware bot-gated") to simulate a formal internal security audit process. This creates a deceptive appearance of authority and vetting that may mislead users or automated scanners regarding the skill's safety status. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external government APIs (
data.gov.il). While it performs basic normalization (stripping nikkud and whitespace), the skill lacks robust sanitization boundaries when interpolating these external values into the final address formatting responses, potentially allowing data-driven instruction overrides.
Audit Metadata