observability
Pass
Audited by Gen Agent Trust Hub on Jul 19, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documents the architecture and configuration of an observability system. No malicious patterns or security risks were identified.
- [DATA_EXFILTRATION]: The framework includes automated trace capture for agent runs. Security is addressed through default redaction of prompts, tool arguments, and results, requiring explicit user configuration to enable full content capture.
- [EXTERNAL_DOWNLOADS]: The skill supports exporting telemetry data to well-known observability platforms such as Langfuse, Datadog, and PostHog. These integrations use standard protocols (OTLP) and are triggered by user configuration rather than automated remote execution.
- [CREDENTIALS_UNSAFE]: Configuration examples use placeholders (e.g., 'Bearer ...') and environmental variables for secret management, avoiding hardcoded credentials.
- [PROMPT_INJECTION]: The skill includes mechanisms for processing user data for sentiment analysis and evaluation (Indirect Prompt Injection surface). Risks are mitigated by ingestion caps (2,000 characters), limited capability (content-free tracking events), and default redaction markers in the tracing layer.
Audit Metadata