ai-observability-langfuse

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides documentation and examples for LLM observability using Langfuse. No malicious patterns were detected.\n- [CREDENTIALS_UNSAFE]: The skill correctly instructs users to use environment variables (LANGFUSE_SECRET_KEY, LANGFUSE_PUBLIC_KEY) rather than hardcoding credentials. Examples use clear placeholders (e.g., sk-lf-...).\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data (LLM inputs and outputs) into the Langfuse observability platform.\n
  • Ingestion points: LLM call inputs and outputs are captured via span.update() calls in examples/tracing.md and automated wrappers in examples/openai-integration.md.\n
  • Boundary markers: Explicit boundary markers for the data stored within Langfuse traces are not defined in the provided examples.\n
  • Capability inventory: The skill uses network operations to transmit captured data to the Langfuse API.\n
  • Sanitization: The skill includes a pattern for PII masking in examples/core.md using regex-based redaction (e.g., credit card and email patterns) before data transmission.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 03:56 PM
Security Audit — agent-trust-hub — ai-observability-langfuse