[FEATURE] Flag to limit amount of turn history sent

Resolved 💬 5 comments Opened Apr 1, 2026 by aaron-baff-ad-net Closed May 12, 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

I'm trying to use -p mode for a relatively long running process. Taking in a prompt, making multiple tool calls, one of which is to get work from a queue to process on. These work are generally unrelated to anything else, and so I don't need the entire turn history to be sent back and cost me tokens every time.

This ends up KILLING my subscription session token usage, and I'm on Max plan.

Proposed Solution

Not sure how exactly, but have some kind of flag to limit the maximum number of previous turn history that is sent back as part of the invocation.

Alternative Solutions

Only alternative is, I think, to specifically have claude -p exit after every single task, and pay a penalty in startup/re-processing the initial prompt.

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

I have a non-trivial system prompt that is in a file, with instructions as to the MCP to use, what tool to start using at startup, what the work item(s) are, what to do with them, and how to discover more work to end up enqueuing to then continue processing at a later time.

Additional Context

_No response_

View original on GitHub ↗

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