skills/jeremylongshore/tons-of-skills-marketplace/juicebox-reference-architecture/Gen Agent Trust Hub
juicebox-reference-architecture
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes job descriptions and candidate profile data that could contain malicious instructions designed to influence the agent's scoring or export logic.
- Ingestion points: External data enters the system through SearchCriteria and profile objects returned from the Juicebox API.
- Boundary markers: Explicit delimiters or 'ignore' instructions for the AI scoring phase are not present in the reference implementation.
- Capability inventory: The architecture includes network writes to Juicebox and Merge APIs, database writes to PostgreSQL, and caching in Redis.
- Sanitization: While the skill instructions mandate the use of synthetic data and redacted telemetry in sandbox environments, the reference code does not show character escaping or content filtering for data processed by the AI models.
- [EXTERNAL_DOWNLOADS]: The service layer communicates with external recruitment APIs to perform its primary function.
- Fetches candidate search and enrichment data from Juicebox AI's official API.
- Synchronizes candidate data with external ATS platforms using the Merge API.
- These interactions target well-known recruitment service domains consistent with the skill's purpose.
Audit Metadata