[BUG]

Resolved 💬 2 comments Opened Jun 13, 2026 by joonaasst Closed Jun 17, 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?

Title: Transcript not durably flushed on host crash + resume/--continue reattaches to wrong (older) session

Environment: Claude Code CLI, Windows 11, Windows Terminal + PowerShell host, launched with --remote-control and --continue.

Two related issues observed after the terminal/host process crashed hard:

1) Lost transcript tail (no durable flush)

  • A session ran ~50 min and produced real work (git commits), then the PowerShell host crashed.
  • After the crash, NO .jsonl under ~/.claude/projects/<project>/ contained any entries from that session's time window. The entire tail appears to have lived only in the process buffer and was never written to disk.
  • Expected: each appended message is durably flushed (append + fsync) so a crash loses seconds, not the whole session.

2) Resume / --continue reattaches to the wrong (older) session

  • claude --continue and the picker's top entry both resolve to the most-recently-written transcript in the project dir. Because the session was repeatedly --continue'd, ONE transcript grew across 3 days (~12 MB) and every relaunch reattached to it instead of starting fresh.
  • After a crash, choosing "continue" / the top picker entry drops you into this old long-lived session, not the one you were actually in — the crashed session is effectively unrecoverable.
  • The --remote-control NAME attached to that single file changed across resumes, so in the picker the same physical transcript shows under several different names, making sessions impossible to tell apart.

Suggested fixes:

  • Durable per-message flush of the transcript JSONL.
  • Show session metadata in the picker (last-activity time, size, message count, cwd) so sessions are distinguishable.
  • Stable session identity across resumes; consider auto-rotating/splitting very large or multi-day transcripts so one file doesn't grow unbounded.

What Should Happen?

find the right session

Error Messages/Logs

Steps to Reproduce

Title: Transcript not durably flushed on host crash + resume/--continue reattaches to wrong (older) session

Environment: Claude Code CLI, Windows 11, Windows Terminal + PowerShell host, launched with --remote-control and --continue.

Two related issues observed after the terminal/host process crashed hard:

1) Lost transcript tail (no durable flush)

  • A session ran ~50 min and produced real work (git commits), then the PowerShell host crashed.
  • After the crash, NO .jsonl under ~/.claude/projects/<project>/ contained any entries from that session's time window. The entire tail appears to have lived only in the process buffer and was never written to disk.
  • Expected: each appended message is durably flushed (append + fsync) so a crash loses seconds, not the whole session.

2) Resume / --continue reattaches to the wrong (older) session

  • claude --continue and the picker's top entry both resolve to the most-recently-written transcript in the project dir. Because the session was repeatedly --continue'd, ONE transcript grew across 3 days (~12 MB) and every relaunch reattached to it instead of starting fresh.
  • After a crash, choosing "continue" / the top picker entry drops you into this old long-lived session, not the one you were actually in — the crashed session is effectively unrecoverable.
  • The --remote-control NAME attached to that single file changed across resumes, so in the picker the same physical transcript shows under several different names, making sessions impossible to tell apart.

Suggested fixes:

  • Durable per-message flush of the transcript JSONL.
  • Show session metadata in the picker (last-activity time, size, message count, cwd) so sessions are distinguishable.
  • Stable session identity across resumes; consider auto-rotating/splitting very large or multi-day transcripts so one file doesn't grow unbounded.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v1.12603.1

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

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