agent-observability

Pass

Audited by Gen Agent Trust Hub on Sep 24, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from model responses and tool outputs to generate logs and traces, creating a potential ingestion surface for indirect prompt injection.
  • Ingestion points: The tracing wrappers in SKILL.md (traced_llm_call, traced_tool) capture raw message content, tool arguments, and model responses.
  • Boundary markers: The implementation does not explicitly utilize prompt delimiters or 'ignore' instructions for the data being monitored, though this is typical for observability-focused skills.
  • Capability inventory: The skill possesses network capabilities via the OpenTelemetry OTLP exporter and httpx for context propagation, along with structured logging to standard output.
  • Sanitization: The skill mitigates risks by including a comprehensive PII redaction module (pii_redactor.py). This module uses regular expressions to scrub sensitive patterns including emails, SSNs, credit cards, JWTs, and API keys from all spans and logs before they are exported to external systems.
  • [SAFE]: The skill is designed with a strong security posture. It includes specific mechanisms for cost monitoring to prevent resource exhaustion (runaway loops) and implements data privacy controls (redaction) as a core feature rather than an afterthought.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 24, 2026, 01:40 AM
Security Audit — agent-trust-hub — agent-observability