[Feature Request] Warn on large session transcript + show size/health indicator in session list

Open 💬 1 comment Opened Jul 1, 2026 by andresmassello

Problem

Long-running sessions accumulate a large .jsonl transcript on disk. Once a
transcript gets big enough (in my case ~80 MB), the desktop app becomes
unusable — it freezes/hangs when opening the session, with no prior warning.
I lost access to a pinned, important session this way. The transcript is safe on
disk, but the app can no longer open it, and nothing warned me it was heading
there.

Environment:

  • Claude Code desktop app on Windows 10
  • Session transcript: ~/.claude/projects/<project>/<id>.jsonl reached ~80 MB
  • No warning shown at any point before the hang

This is distinct from context-window compaction (#19567). The context window is
capped and auto-compacts; the problem here is the on-disk transcript / client
rendering
, which grows unbounded and eventually hangs the UI.

Requested features (two, ideally independent)

1. In-session warning before the transcript gets dangerous

When the active session's transcript crosses a size threshold (e.g. a soft
warning at ~25 MB and a hard alert at ~50–60 MB), surface a non-blocking notice
suggesting the user start a fresh session. Ideally with a configurable threshold.

Today the only workaround is a custom hook on UserPromptSubmit that stats the
transcript file and emits a systemMessage. That works, but it's the kind of
safety net that should be built in.

2. Size / health indicator in the session list (sidebar & /resume picker)

Show each session's transcript size, with a simple traffic-light (green /
yellow / red) so users can see at a glance which sessions are at risk of
hanging — before clicking into one and freezing the app.

Why it matters

Users pin and rely on important sessions. Losing one silently, with no warning
and no in-app indicator, is a bad surprise — and the data is right there on disk,
just unopenable. A warning + a visible indicator would prevent the loss entirely.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗