[FEATURE] Ability (command?) to tell Claude it is now working on a new task, but keep relevant context knowledge.

Resolved 💬 5 comments Opened Dec 3, 2025 by ANogin Closed Feb 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

Often I tell Claude to do a task ("X"). After it is done, I ask it to do a subsequent task "Y" without /clear because the new task is highly related and is further work on the same part of the code, so I want it to keep existing code knowledge in the context. However Claude is now thinking it was tasked to do "X+Y" and keeps dragging X into the discussion in various ways:

  • Plan is often contaminated by X (see e.g. #12505) whereas I want it to keep the old plan file alone (but not overwrite - what if I want to refer back to it later if I need to revisit X), and start a new one.
  • When it summarizes what it has done in its message to me before stopping, it will summarize both X and Y work, unnecessarily spending tokens on X
  • When it later compacts the context, it will include a lot of details about the X work, rather than only including things that are relevant for Y.

Proposed Solution

I do not feel I have enough expertise to know what's the best way to do it, but theoretically either of two ways:

  • A command that makes it clear to Claude what is going on
  • A specialized version of /compact that tells it to preserve the general codebase knowledge (maybe more detailed than regular /compact), but none of the task-specific knowledge.

I am guessing there would be an input token count vs output token count tradeoff between the two.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

_No response_

Additional Context

_No response_

View original on GitHub ↗

This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗