A Watch/clock for ClaudeAi and Code (quick win)

Resolved 💬 3 comments Opened May 1, 2026 by ktkorpela-prog Closed May 4, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Claude Code has no reliable sense of actual time. It infers session state from conversation length rather than from the timestamps on messages. The result is wrap-up suggestions twenty minutes into a fresh morning session, prompts to "call it a day" that have nothing to do with the time of day, and a model behaving as if it were late evening when it is in fact 9am.

Proposed Solution

Give Claude Code access to the timestamp on each incoming prompt, so it can read the actual time and the gap since the last message rather than only the count of messages in the conversation buffer. A simple watch the model can check, surfaced into context the same way the working directory or git state already is.

Alternative Solutions

_No response_

Priority

Low - Nice to have

Feature Category

CLI commands and flags

Use Case Example

I open a fresh session most mornings. Within twenty or thirty minutes, Claude Code regularly suggests we wrap up or continue another time. I am at the start of my workday. The model is reading session length as elapsed time, which is wrong almost every morning. My current workaround is a manual good night / good morning ritual that gives the model a session boundary it cannot generate on its own. It works. But the ritual exists because the model cannot tell what time it is.

Additional Context

I posted about this on Substack and it resonated with other users, including one who tested it across models and found Haiku two hours off. Time is the kind of context where freshness matters more than presence. A model that thinks it is 11pm when it is 9am weights responses differently and behaves inside an imagined fatigue state that does not match the real one.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗