Feature request: optional timestamps on assistant responses in long-running sessions
The problem
In long-running sessions (days of supervision work, background agents reporting in, a user checking from their phone), the transcript gives no sense of when anything was said. Scrolling back through a day of updates, there is no way to tell whether a status report is from five minutes ago or five hours ago without cross-referencing something external.
The request
An option (setting or slash command) that prefixes each assistant response with the local time it was written, e.g. [28 Aug 21:04]. Nothing fancier than that.
Workaround today
A CLAUDE.md instruction telling the model to start every response with date output works, but it costs a tool call or risks a stale guess, and every user has to reinvent it. The harness knows the real time for free.
Came out of a multi-day session supervising a CI/agent pipeline, where "how's it going" answers from different hours look identical in the scrollback.
🤖 Generated with Claude Code