opik-evaluate

Pass

Audited by Gen Agent Trust Hub on Aug 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions involve processing external data sources such as pipeline traces, retrieved document chunks, and user-provided test cases. This creates an inherent surface for indirect prompt injection where malicious instructions embedded in the analyzed data could potentially influence the agent's evaluation results. \n
  • Ingestion points: Traces are analyzed in references/error-analysis.md, document chunks are processed in references/evaluate-rag.md, and test suite items are defined in SKILL.md. \n
  • Boundary markers: The skill suggests structured prompt templates for LLM judges (e.g., in references/write-judge-prompt.md), though it does not explicitly mandate the use of high-strength delimiters to isolate untrusted data. \n
  • Capability inventory: The skill utilizes the Opik SDK to execute evaluation tasks, which may include running user-defined logic or functions. \n
  • Sanitization: The instructions focus on evaluation logic and do not provide specific guidance on sanitizing or filtering input data for potential injection attacks.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 29, 2026, 01:56 AM
Security Audit — agent-trust-hub — opik-evaluate