Problem
session show --tail N prints raw JSONL — fine for a quick check, useless for actually reading a run: what the agent did, what tools it called, where it went wrong.
Proposal
tracecraft replay build <session-id> --output run.html — fetch a mirrored session's parts and render a single self-contained HTML file: message timeline, tool calls collapsed/expandable, timestamps, and the session metadata header. No server, no JS dependencies fetched at view time — one file you can open, archive, or attach to an incident report.
Pairs with mirroring: mirror on every machine, replay from any machine.
Problem
session show --tail Nprints raw JSONL — fine for a quick check, useless for actually reading a run: what the agent did, what tools it called, where it went wrong.Proposal
tracecraft replay build <session-id> --output run.html— fetch a mirrored session's parts and render a single self-contained HTML file: message timeline, tool calls collapsed/expandable, timestamps, and the session metadata header. No server, no JS dependencies fetched at view time — one file you can open, archive, or attach to an incident report.Pairs with mirroring: mirror on every machine, replay from any machine.