Multiple Cowork sessions lost conversation transcripts (transcriptUnavailable: true)

---
type: bug-report
created: 2026-07-08
status: ready-to-submit
---

Bug Report: Multiple Claude Cowork sessions lost conversation transcripts

Summary

Multiple local Claude Code / Claude Cowork sessions on my desktop app show up in the session list with their correct title and metadata, but the actual conversation content (transcript) is gone. Opening these sessions shows "Session not found on disk" or an empty conversation. This affected real business work — strategy discussions, planning, and content drafting that I had typed directly into these threads.

Environment

• App: Claude Cowork (desktop app)
• OS: Windows 11 Pro
• Model used across affected sessions: claude-sonnet-4-6

What I found

The app's own local session state files (local_*.json under AppData\Roaming\Claude\claude-code-sessions\) confirm this is not a display bug — each affected session has its own field explicitly set:

"transcriptUnavailable": true

The session metadata (title, creation date, last activity, permission settings) is intact, but the underlying transcript is confirmed missing at the app level, not just missing from the UI.

Affected sessions (9 confirmed via transcriptUnavailable: true)

| Date created | Completed turns | Session title |
|---|---|---|
| 2026-04-21 | 33 | Code Migration/Training |
| 2026-04-22 | 23 | Email Triage |
| 2026-04-22 | 1 | Comprehensive Wholesale Strategy |
| 2026-04-28 | 13 | Claude Code Setup |
| 2026-04-29 | 15 | Beyond The StEPS Scripting |
| 2026-04-29 | 6 | Draft defective socks complaint email to InnovaKnits |
| 2026-05-19 | 15 | Obsidian Training |
| 2026-05-19 | 7 | Apolla Shirt Design |
| 2026-05-19 | 13 | Wholesale Email |

All occurred within a roughly one-month window (April 21 – May 19, 2026). Several (like "Beyond The StEPS Scripting" and "Wholesale Email") had substantial conversations (13-15 completed turns) with real strategic content that is now unrecoverable on my end.

Notable detail

Both sessions I inspected in depth had "autoArchiveExempt": true set — meaning the one setting that appears designed to protect a session from archiving-related data loss did not prevent this from happening. This suggests the bug may be in how transcripts are written/finalized (possibly during an archive or compaction step), not specifically in the auto-archive feature itself.

Additional related issue (possibly separate root cause)

Two other threads I was looking for ("Wholesale - Nursing" and "Faire") have no local session file at all — not even a metadata stub with transcriptUnavailable: true. These may have been claude.ai web sessions (hosted) rather than local Claude Code sessions, opened via browser tabs inside the desktop app. If so, this points to a second, separate storage layer (hosted claude.ai chat state) that can also lose data independently of the local session store. I'm flagging this in case it's relevant, though I was not able to confirm the exact mechanism from my end.

Impact

This resulted in permanent loss of real business strategy content (wholesale channel strategy, a scripting project, and email drafting work), with no ability to recover it from any location I have access to (checked: local session files, transcript files, Windows Recycle Bin, and confirmed no antivirus/cloud-sync interference on my machine).

Request

Could you help identify:

1. What causes transcriptUnavailable: true to get set on an otherwise-intact session record — is this a write failure, a crash during a save/archive step, or something else?
2. Whether there's a way to prevent this going forward (e.g., a setting, or a fix in an upcoming release)?
3. Whether the hosted claude.ai session loss (Faire example) is related or a separate issue worth reporting elsewhere?

Happy to provide additional local file details if it helps reproduce or diagnose this.