[FEATURE] Single-prompt context scoping (`/once` or `--no-history` flag)

Resolved 💬 2 comments Opened Mar 4, 2026 by simaximum Closed Mar 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

When deep in a long conversation, there are frequent cases where you want to ask a quick, self-contained question — a small detail, a one-off lookup, a sanity check — without the model needing to process (and pay for) the entire conversation history.

Currently, the only options are:

  • /clear — too destructive, loses all context
  • /compact — preserves context but requires an LLM call, adds latency and cost
  • Neither option keeps you "in" the current session for follow-up

Proposed Solution

A mechanism to scope a single prompt to a reduced context — ideally just the last exchange or a user-defined window — without modifying the ongoing session history.

Possible interfaces:

/once What does this error mean?

or a per-message flag:

>>> --fresh What does this error mean?

Behavior:

  • Claude answers using only the last N exchanges (or no history at all)
  • The session history is not modified — the next prompt resumes full context as usual
  • The response is appended to session history normally

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

You're mid-session on a complex refactor. You want to quickly ask what a specific bash flag does, or check a regex, or clarify a concept. You don't want to burn 150K tokens of context on a question that needs zero prior context. And you don't want to break your session flow.

Additional Context

Related

  • Similar to how subagents get a fresh context window, but inline within the current conversation
  • Complements /compact and /clear rather than replacing them

Expected Benefit

  • Reduced token cost for simple follow-up questions
  • Faster responses (smaller context = less latency)
  • No disruption to the ongoing session

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗