Temporal Attack Pattern Detection in Multi-Agent AI Workflows: An Open Framework for Training Trace-Based Security Models

This project addresses a critical and emerging gap: security for autonomous agent workflows where traditional request-response filtering is insufficient. Its primary value lies in the curated dataset (80k+ examples) and the decision to use OpenTelemetry (OTel) as the data source, which aligns with industry-standard observability practices. However, the defensibility is low (3/10) because the project currently lacks community adoption (0 stars) and the 'moat' consists primarily of a fine-tuning recipe and a static dataset rather than a protected network effect or proprietary engine. Large observability players like Datadog, New Relic, or Splunk are the natural owners of this space; they could easily integrate similar LLM-based trace analysis into their existing security products (SIEM/APM). Furthermore, as frontier labs (OpenAI/Anthropic) release their own agent orchestration layers, they are likely to build native security monitoring that renders external third-party 'trace-checkers' less necessary for the average user. The specific focus on ARM64 hardware is a notable technical detail but does not provide a competitive advantage in a cloud-dominant landscape. This is a high-quality research contribution that likely serves as a blueprint for others rather than a standalone commercial entity.