implementing-observability

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFECREDENTIALS_UNSAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: Hardcoded default credentials are present in the observability stack configuration and documentation for local development.
  • Evidence: GF_SECURITY_ADMIN_PASSWORD=admin in examples/lgtm-docker-compose/docker-compose.yml and references/lgtm-stack.md.
  • Evidence: Documentation in multiple README files mentions the admin/admin default credential for Grafana access.
  • [DYNAMIC_EXECUTION]: Utility scripts are used to generate application code and configuration files from internal templates at runtime.
  • Evidence: scripts/setup_otel.py creates instrumentation.py and main_example.py by interpolating variables into Python code templates.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements monitoring and logging infrastructure that ingests external, untrusted data (logs, traces, metrics), creating a surface for indirect injection if processed by an agent.
  • Ingestion points: Loki (logs), Tempo (traces), and Mimir (metrics) collection services.
  • Boundary markers: Not explicitly configured in the provided configuration templates.
  • Capability inventory: The skill provides scripts for querying and validating telemetry data.
  • Sanitization: No explicit sanitization or filtering logic is implemented for ingested telemetry content.
  • [COMMAND_EXECUTION]: The validate_metrics.py script performs network requests to user-defined endpoints which could be misused for internal network scanning (SSRF).
  • Evidence: requests.get(endpoint) in scripts/validate_metrics.py where endpoint is a command-line argument provided at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 09:34 AM
Security Audit — agent-trust-hub — implementing-observability