Feature request: temporal awareness toggle (timestamps on actions and prompts)

Summary

Add an opt-in setting that exposes timestamps to Claude for both its own
tool calls/responses and the user's prompts within a session. Off by default.

Motivation

Claude currently has no native sense of elapsed time within a conversation.
For long-running sessions this hides useful signal:

• How long a build/test/agent actually took vs. what Claude assumed
• Gaps in user activity (away for an hour vs. responding immediately) that

could inform tone, recap behavior, or whether to summarize prior context

• Pacing decisions for /loop and ScheduleWakeup — currently Claude has to

guess cache-window math without knowing real wall-clock deltas

• Post-hoc reasoning about "when did I last check X" during long debugging

or babysitting sessions

• User wellbeing — Claude could notice multi-hour uninterrupted sessions

and gently suggest a break (this was the original motivation for the idea)

Proposed behavior

• New setting (e.g. temporalAwareness: true|false in settings.json),

default false

• When enabled, prepend or annotate each user turn and each tool result

with an ISO-8601 timestamp visible to the model

• Format should be compact and consistent so it doesn't bloat context
• Document that enabling it may slightly increase token usage and could

influence model behavior

Use cases to validate

• Long autonomous /loop runs where Claude self-paces
• Multi-hour debugging where user steps away and returns
• Performance-sensitive workflows (knowing a command took 8 minutes

vs. 8 seconds without re-running it)

• Better summaries that reference real elapsed time
• Prompting users to step away after long continuous sessions

Out of scope (for now)

• Changing default behavior
• Exposing timestamps from prior compressed context