AI coding agents can read source files, invoke MCP server tools, execute commands and delegate work to subagents. But their chat transcript does not reveal the complete story. Did the agent load the expected skill? Which MCP tool did it call, with which arguments? Was the call part of the main turn or a parallel subagent? How long did each step take and at what cost? GitHub Copilot , Cursor and Claude Code expose supported hooks around sessions, prompts, tools, files, MCP calls, compaction and subagents. In this demo-driven session, we will use them to build a passive flight recorder in .NET. You will then see how to correlate session, turn, tool and subagent identifiers to reconstruct the agent’s activity and summarize tool, MCP and skill usage. No traffic interception, undocumented APIs or vendor marketing: just supported hooks, working code and honest observability boundaries.