See what happened in an agent session.
Replay a Claude Code transcript, inspect tool calls, and trace changes in context.
Or drop a file here — .jsonl or .tape.json
Several at once is fine. The agent-*.jsonl files beside a session are attached to the calls that delegated to them.
Files are read in your browser, not uploaded.
The demo is a fictional session with two tool failures, a context jump, a compaction and a delegated run in it.
Your own transcripts are already on disk, at ~/.claude/projects. Nothing needs to be instrumented before a run.