[BUG] "Prompt is too long" in a new session when another session has an open plan

Status Open
Reported on v2.1.220
Maintainer reply None cached
Activity 0 comments · opened Aug 20, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

A session left waiting at a plan-mode approval prompt causes an
unrelated session to fail with "Prompt is too long".

I had session A open at a presented plan, awaiting accept/reject. In a
separate session B, I asked Claude Code to read a ~78KB text file
(roughly 20k tokens). Session B returned "Prompt is too long".

The file is nowhere near any context limit. I spent significant time
compressing the log file — down to 78KB and then to 14k-token chunks —
before realizing the file size was never the problem. The pending plan
in session A was.

The error message is the core issue: it names a context-length limit,
which points at the input I just supplied. Nothing indicates that state
in a different session is responsible, so there is no path from the
error to the actual cause.

What Should Happen?

Session B's available context should not be affected by an unresolved
plan sitting in session A. Reading a 78KB file in a fresh session
should succeed.

If sessions do legitimately share some budget, the error should say so
— e.g. naming the other session or the pending plan — rather than
reporting a generic prompt-length error that misdirects toward the
file being read.

Error Messages/Logs

Steps to Reproduce

  1. Start session A. Trigger plan mode and let Claude present a plan.

Leave it awaiting approval; do not accept or reject.

  1. Start session B in a separate terminal.
  2. In session B, ask Claude Code to read a moderately large text file

(~78KB / ~20k tokens in my case).

  1. Session B fails with "Prompt is too long".
  2. Resolve or exit the plan in session A, then retry step 3 in

session B.

Environment:

  • Claude Code version: [claude --version]
  • OS / terminal: [e.g. Windows 11, PowerShell / WSL2 Ubuntu]
  • Model: [from /status]

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.220 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

_No response_

View original on GitHub ↗