jev-orchestrator
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external, untrusted data to perform semantic judgments, creating a potential surface for indirect prompt injection.
- Ingestion points: The skill ingests free-text customer requests, student answers, and retrieved RAG passages as described in SKILL.md, references/examples/grading.md, and references/examples/rag.md.
- Boundary markers: Although SKILL.md advises using "minimum relevant state" and "named JSON fields," it does not explicitly require the use of strong delimiters or instructions to ignore embedded prompts within the external data.
- Capability inventory: Decisions produced by the Jev runtime directly determine critical execution paths, such as support ticket routing, tool invocation, and model orchestration.
- Sanitization: There is no explicit requirement or logic for sanitizing or escaping the input text before it is evaluated by the judgment runtime.
Audit Metadata