phoenix-observability
Pass
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install official packages for the Phoenix observability platform and OpenInference instrumentation (e.g., "arize-phoenix", "openinference-instrumentation-openai") and utilize Docker images ("arizephoenix/phoenix"). These are well-known resources from established technology providers.
- [COMMAND_EXECUTION]: Provides standard operational commands for server management ("phoenix serve"), package installation ("pip install"), and containerized deployment ("docker-compose").
- [DATA_EXFILTRATION]: The core functionality of the skill involves collecting and transmitting application trace data, including model inputs and outputs, to a central collector. This is the intended purpose of observability tooling. Users should ensure they use secure and authorized collector endpoints to maintain data privacy.
- [PROMPT_INJECTION]: As an observability platform that processes data from external LLM applications, there is an inherent surface for indirect prompt injection if the ingested data contains adversarial instructions designed to influence evaluation results or display behavior.
- Ingestion points: Data enters the system from external LLM calls via the "Client" API (e.g., "get_spans_dataframe") or OTLP collector endpoints.
- Boundary markers: None explicitly defined in the provided evaluation code snippets.
- Capability inventory: The skill facilitates shell command execution for server operations and network access for trace transmission.
- Sanitization: Standard sanitization for trace data is expected but not explicitly detailed in the documentation.
Audit Metadata