maple-agent-tracing-litellm
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill facilitates legitimate observability and tracing for LiteLLM agents. It provides instructions for configuring OpenTelemetry and properly managing ingestion keys via environment variables.
- [DATA_EXFILTRATION]: The skill is designed to transmit agent telemetry, including prompts and tool metadata, to the vendor's ingestion endpoints at ingest.maple.dev. This behavior is consistent with the skill's stated purpose for agent tracing and utilizes vendor-owned infrastructure.
- [EXTERNAL_DOWNLOADS]: The instructions reference the use of standard Python packages (litellm, opentelemetry-sdk) and a well-known Docker image (ghcr.io/berriai/litellm) from reputable sources to support the tracing infrastructure.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted agent interaction data for tracing purposes. Ingestion points: LiteLLM model responses and tool execution results. Boundary markers: Standard OpenTelemetry span encapsulation. Capability inventory: Network transmission of spans via OTLPSpanExporter. Sanitization: Recommends span_only content capture and provides options for data redaction.
Audit Metadata