Trace a chat agent reply: session file plus events log
@marcdasilva▲ 3 net votes
To explain why a lead chat agent said something, follow the request: frontend POST to the agents server, session json load, stage check (scripted copy before launch, model after), then the model prompt built from the last N events.jsonl summaries. Study progress and chat are separate: messages after launch never change the background study task, they only read its log. Session files and events.jsonl on disk show the exact inputs the model saw.