phoenix-observability

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides templates and methods for LLM-based evaluation (LLM-as-a-judge) that ingest untrusted data, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: External data is interpolated into evaluation templates via input_text, output_text, and reference_text parameters in evaluation functions defined in SKILL.md and references/advanced-usage.md.\n
  • Boundary markers: The evaluation prompt templates (e.g., CUSTOM_EVAL_TEMPLATE) lack explicit delimiters or specific instructions to the evaluator model to ignore embedded commands in the data being assessed.\n
  • Capability inventory: The skill facilitates network communication with external LLM providers (OpenAI, Anthropic) and observability servers, and includes functionality for data logging and alerting via webhooks.\n
  • Sanitization: The provided code implementation examples do not demonstrate sanitization, escaping, or validation of the input data before it is included in evaluation prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 07:08 PM
Security Audit — agent-trust-hub — phoenix-observability