You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug] /resume loads incomplete session context despite full file on disk #22614
I've now reproduced the issue with precise measurements:
Test performed today (Feb 2):
Worked session to 97% context (194.38k/200k tokens, 606 intents)
Used /exit
Used /resume to return to same session
Context dropped to 11% (22.84k/200k tokens)
Assistant had no memory of conversation ("What are we working on?")
Session file verified:
File size: 6.75MB (complete)
866+ message lines in JSONL
No truncation, full content preserved on disk
Only summary entry: "User Exit Session" at end
Key finding: The session file contains the full conversation, but /resume only loads ~11% of it. This is consistent across multiple large sessions (I have 44
sessions totaling 80MB in ~/.claude/projects/).
This appears to be a resume loading issue, not a storage issue. Happy to provide session files or additional logs if helpful.
This might be related to the session file being too large for full context loading — Claude loads what it can and drops the rest.
Cozempic can help by pruning redundant data from the session file before resume, so more of the actual conversation fits in context:
pip install cozempic
cozempic current --diagnose # See what's taking up space
cozempic treat current -rx standard --execute # Prune bloat, keep conversation
It removes progress ticks (40-48% of typical files), stale file reads, oversized tool outputs, and duplicate documents — while preserving the actual conversation DAG. This means more of your real session content survives the resume.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.
Activity
junaidtitan commentedon Feb 6, 2026
This might be related to the session file being too large for full context loading — Claude loads what it can and drops the rest.
Cozempic can help by pruning redundant data from the session file before resume, so more of the actual conversation fits in context:
It removes progress ticks (40-48% of typical files), stale file reads, oversized tool outputs, and duplicate documents — while preserving the actual conversation DAG. This means more of your real session content survives the resume.
github-actions commentedon Mar 6, 2026
Closing for now — inactive for too long. Please open a new issue if this is still relevant.
github-actions commentedon Mar 16, 2026
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.