ai-observability-langfuse
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill explicitly instructs users to manage sensitive API keys and base URLs using environment variables (e.g.,
LANGFUSE_SECRET_KEY), preventing the risk of hardcoded credentials in source code.\n- [SAFE]: Guidance is provided for implementing data masking on theLangfuseSpanProcessorto redact sensitive PII such as credit card numbers and email addresses before data is transmitted to the observability platform.\n- [SAFE]: The documented packages and dependencies (Langfuse and OpenTelemetry) are official, well-known libraries from trusted organizations, directly relevant to the skill's functional purpose.\n- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing external LLM data for tracing and evaluation.\n - Ingestion points: Untrusted data enters the context through
langfuse.prompt.get(),dataset.runExperiment(), and the automated capture of LLM inputs/outputs viaobserveOpenAI().\n - Boundary markers: No specific boundary markers or instruction-bypass warnings are implemented, as the skill acts as a passive observation layer.\n
- Capability inventory: The skill's capabilities are limited to network communication with the Langfuse API for exporting spans, scores, and dataset results.\n
- Sanitization: The skill provides PII masking functionality (regex-based redaction) for data at rest, but does not perform active sanitization of prompts or outputs for malicious instructions.
Audit Metadata