agent-observability-eval-pipeline

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). Phase 1 runs a Datadog trace search (search_llmobs_spans) and then Phase 4 samples production trace content into a dataset via the sub-skills (agent-observability-session-classify/agent-observability-eval-bootstrap --emit-dataset), which means outsider-authored free text from spans (e.g., prompts/responses/tools recorded in traces) can be ingested at runtime.