Encrypted agent-to-agent traffic of an LLM multi-agent system: TLS 1.3 A2A captures and features (collection C1)

Packet captures and features of encrypted agent-to-agent (A2A) traffic from an LLM multi-agent system, released with the paper "What TLS leaves on the wire: inferring agent roles, topology and workflows of LLM multi-agent systems from encrypted agent-to-agent traffic" (More, Pujari and Berhe, EDI 2027, Procedia Computer Science). A four-agent A2A system (orchestrator, executor, retriever, validator; a2a-sdk 0.3.26, Llama 3.2 3B via Ollama) ran in three topologies (star, chain, mesh) on four workflows (code review, data analysis, research retrieval, support triage). Collection C1 holds 600 tasks (50 per topology and workflow) captured in one interleaved session with TLS 1.3 on every agent link. Captures are cut to link, IP and TCP headers (timestamps and on-wire lengths kept; no payload bytes) and come with JSON label sidecars. Features: per-connection and per-trace feature matrices (.npz) with labels.json. All traffic was generated by the authors on their own machines; no third-party traffic is included. Code: https://github.com/MayureshMore/Where-the-Fingerprint-Lives-A-Passive-Metadata-Measurement-of-Leakage-in-LLM-Multi-Agent-A2A-Syst