## Description When a long session is compacted (via /compact) and then resumed later (via --resume or /resume picker), the resumed session loads from a late checkpoint instead of the compaction point. This causes: 1. **Truncated exports** — /export from the resumed session produces a fraction of the content vs /export from the original running session (527 lines vs 8,902 lines for the same session) 2. **Lost context for Claude** — Claude operates with partial conversation history after resume

Resume after compaction loses conversation content — exports truncated

Status Fixed / completed

Maintainer reply None cached

Activity 4 comments · opened Apr 3, 2026 · closed Apr 3, 2026

Description

When a long session is compacted (via /compact) and then resumed later (via --resume or /resume picker), the resumed session loads from a late checkpoint instead of the compaction point. This causes:

Truncated exports — /export from the resumed session produces a fraction of the content vs /export from the original running session (527 lines vs 8,902 lines for the same session)
Lost context for Claude — Claude operates with partial conversation history after resume, degrading response quality
Non-deterministic checkpoint selection — the same session ID resumes from different points depending on when you resume

Reproduction

Start a session, work for several hours (4,000+ messages)
/compact mid-session
Continue working (adds ~2,000 more messages)
/export from the running session → complete export (8,902 lines)
Exit
claude --resume <session-id>
/export from resumed session → truncated export (527 lines, starts from a random late point)

Expected

Resume should load from the compaction point (or the full session), and exports should be identical regardless of whether exported from the original or resumed session.

Environment

Claude Code v2.1.91
Opus 4.6 (1M context)
Windows 11 / MSYS2/Git Bash
Session JSONL: 41MB, 6,149 lines, 4,451 messages
Multiple compactions in the same session

View original on GitHub ↗

4 Comments

belumume · 6 months ago

Closing — filed prematurely without proper diagnosis. The resume issue appears to be related to /desktop session transfer, not a general compaction bug. Will refile with accurate root cause if confirmed.

github-actions[bot] · 6 months ago

Found 3 possible duplicate issues:

This issue will be automatically closed as a duplicate in 3 days.

If your issue is a duplicate, please close it and 👍 the existing issue instead
To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

kolkov · 6 months ago

@belumume Don't refile — the root cause is already identified and documented.

Your symptoms (41MB JSONL, truncated resume, 527 vs 8,902 lines) match exactly what we found through source code analysis. It's NOT specific to /desktop session transfer — it's a general v2.1.91 regression in the session loading pipeline.

Three specific code changes cause this:

walkChainBeforeParse removed in v2.1.91 — fork pruning for >5 MB files no longer happens
New ExY timestamp fallback bridges across unrelated fork branches
Missing leafUuids check in getLastSessionLog (line 3900)

Full analysis with exact file:line references from v2.1.88 TypeScript source: #43044

If you need to recover the lost session content now, you can use our Go tool that reads JSONL directly (bypassing the broken resume):

go install github.com/kolkov/ccdiag@latest
ccdiag recover --output full -o recovery.md <your-session.jsonl>

https://github.com/kolkov/ccdiag