Chat history loss after PC restart, /resume recovers only 20-40% of context

Description

Chat history is being lost after PC/session restart. The /resume command fails to fully restore conversation context, recovering only approximately 20-40% of the original chat history. The agent loses all memory of previous conversations.

Timeline

First observed: ~2 weeks ago when latest channel reached version 2.1.29
Persisted through: latest versions 2.1.29 → 2.1.35
Now affects stable: Issue appeared on stable channel starting with version 2.1.25

Steps to Reproduce

Have an active conversation with significant context
Restart PC or terminal session
Use /resume to restore the previous conversation
Observe that 60-80% of the conversation context is missing

Expected Behavior

/resume should fully restore the conversation history and context.

Actual Behavior

Chat history is lost after restart
/resume only recovers 20-40% of the conversation
Agent has no memory of previous discussion topics

Environment

Claude Code version: 2.1.25 (stable)
OS: Ubuntu 25.10
Kernel: Linux 6.17.0-12-generic x86_64
Desktop: GNOME Shell 49.0 (Wayland)

My colleague is experiencing the same issue on Windows with version 2.1.34

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

Same issue as #22107

This bug is making Claude Code practically unusable for me. I can't take advantage of the new Claude Opus 4.5 model properly because every session restart loses most of the conversation context.

As a result, I'm forced to use Cursor Agent instead for any serious work that requires maintaining context across sessions.

+1 same exact issue. Cursor crashing (I use it only as IDE and have my CC sessions inside its integrated powershell) leaves me unable to resume conversations. I see the screen (tui) flash briefly with my conversation history but then goes back to the empty state. Pressing 'crtl + w' closes that powershell window and doesnt 'show all worktrees'

really bad issue for me as I have ongoing conversations with a lot of time spent on planning.

Additional data: `sessions-index.json` stops being updated

Environment:

Claude Code 2.1.42 (stable)
Windows 11 Pro 10.0.26200
Issue started: February 15, 2026

Observation

I noticed that /resume couldn't find my recent sessions. I compared sessions on disk vs sessions registered in the index and found a discrepancy:

Sessions on disk:  118
Sessions in index: 87

LAST 15 sessions on disk (sorted by mtime):
2026-02-14 22:46 | 318148cd | 1,478KB | IN INDEX
2026-02-15 09:21 | 5d158395 |     5KB | *** MISSING ***
2026-02-15 11:38 | a22bc060 |   526KB | IN INDEX
2026-02-15 11:38 | bf049b7b |   396KB | IN INDEX
2026-02-15 11:38 | c1d91bfa |   454KB | IN INDEX
2026-02-15 11:38 | 0435bc47 |    73KB | IN INDEX
2026-02-15 12:25 | 1d003a66 |   611KB | IN INDEX
2026-02-15 13:13 | 66d486cc |   365KB | *** MISSING ***
2026-02-16 00:08 | 9546067c |19,803KB | *** MISSING ***
2026-02-16 07:48 | 621e03cb |    18KB | *** MISSING ***
2026-02-16 08:03 | 90a2c3f3 | 5,187KB | IN INDEX (created before the issue)

Facts

sessions-index.json was last modified on Feb 15 at 13:08
4 sessions created after that timestamp exist on disk but are missing from the index
These sessions are invisible to /resume despite being fully intact on disk
One of them (9546067c, 20MB) represents several hours of work

Additional finding: manually adding entries doesn't fix `/resume`

I tried manually adding the missing sessions to sessions-index.json with the correct structure (identical keys: sessionId, fullPath, fileMtime, firstPrompt, summary, messageCount, created, modified, gitBranch, projectPath, isSidechain). The entries were successfully added and the JSON file validates correctly (118 entries, all keys matching existing entries).

However, /resume still doesn't show the added sessions, even after restarting Claude Code in a new terminal. The file was not overwritten by Claude Code on restart (still shows our modifications with correct timestamp).

This suggests that either:

/resume does not read sessions-index.json directly (or uses it only as a cache that gets regenerated from another source)
There is an additional mechanism or secondary index that controls which sessions appear in the picker
The index file may be read only once at some initialization point and cached in memory differently

Possible investigation leads

The index file stopped being updated at a specific point in time — something may have caused writes to sessions-index.json to silently fail from that point on
I notice that at 2026-02-15 11:38, 4 sessions were created almost simultaneously — could be worth checking if concurrent index writes are properly handled
The 31 other missing sessions (118 on disk vs 87 in index) suggest this may have happened before as well, or that some session types are never indexed
Manually patching the index file does not restore visibility in /resume — the picker seems to rely on something else than (or in addition to) this file

Diagnostic scripts

Script 1 — Compare disk vs index (adapt <PROJECT> to your project path):

import json, os, datetime

index_path = os.path.expanduser("~/.claude/projects/<PROJECT>/sessions-index.json")
sessions_dir = os.path.dirname(index_path)

with open(index_path, encoding="utf-8") as f:
    indexed_ids = {e["sessionId"] for e in json.load(f).get("entries", [])}

for f in sorted(os.listdir(sessions_dir)):
    if not f.endswith(".jsonl"):
        continue
    sid = f.replace(".jsonl", "")
    fpath = os.path.join(sessions_dir, f)
    mtime = datetime.datetime.fromtimestamp(os.path.getmtime(fpath))
    size = os.path.getsize(fpath) // 1024
    status = "OK" if sid in indexed_ids else "*** MISSING ***"
    print(f"{mtime:%Y-%m-%d %H:%M} | {sid[:12]} | {size:>6}KB | {status}")

Script 2 — Add missing sessions to index (attempted fix, did not restore /resume visibility):

import json, os, datetime, shutil

index_path = os.path.expanduser("~/.claude/projects/<PROJECT>/sessions-index.json")
sessions_dir = os.path.dirname(index_path)

shutil.copy2(index_path, index_path + ".bak")

with open(index_path, encoding="utf-8") as f:
    data = json.load(f)

indexed_ids = {e["sessionId"] for e in data.get("entries", [])}

for fname in os.listdir(sessions_dir):
    if not fname.endswith(".jsonl"):
        continue
    sid = fname.replace(".jsonl", "")
    if sid in indexed_ids:
        continue
    fpath = os.path.join(sessions_dir, fname)
    if os.path.getsize(fpath) == 0:
        continue
    mtime = os.path.getmtime(fpath)
    ctime = os.path.getctime(fpath)
    # Parse first user prompt
    first_prompt = "No prompt"
    msg_count = 0
    try:
        with open(fpath, encoding="utf-8") as sf:
            for line in sf:
                try:
                    msg = json.loads(line.strip())
                    if msg.get("type") in ("user", "assistant"):
                        msg_count += 1
                        if first_prompt == "No prompt" and msg.get("type") == "user":
                            content = msg.get("message", {}).get("content", "")
                            if isinstance(content, list):
                                for c in content:
                                    if isinstance(c, dict) and c.get("type") == "text":
                                        txt = c.get("text", "")
                                        if txt and not txt.startswith("[Request"):
                                            first_prompt = txt[:200]
                                            break
                except json.JSONDecodeError:
                    continue
    except Exception:
        continue
    data["entries"].append({
        "sessionId": sid,
        "fullPath": fpath,
        "fileMtime": int(mtime * 1000),
        "firstPrompt": first_prompt,
        "summary": "",
        "messageCount": msg_count,
        "created": datetime.datetime.fromtimestamp(ctime).strftime("%Y-%m-%dT%H:%M:%S.000Z"),
        "modified": datetime.datetime.fromtimestamp(mtime).strftime("%Y-%m-%dT%H:%M:%S.000Z"),
        "gitBranch": "main",
        "projectPath": "<YOUR_PROJECT_PATH>",
        "isSidechain": False,
    })

with open(index_path, "w", encoding="utf-8") as f:
    json.dump(data, f, ensure_ascii=False, indent=2)

/resume has been broken for over a month. This makes Claude Code unusable for me. I had to build a workaround skill (“/resume-fix”) that searches the conversation history and exports it to Markdown, but it’s still inconvenient. I’m honestly surprised such a critical bug hasn’t been fixed yet. This was the first time in months I switched to other agents, and I ended up downgrading my Claude Code subscription because of it.

I had a similar problem:

Title: Resuming migrated sessions fails due to thinking/redacted_thinking block validation
Problem: After migrating ~/.claude to a new machine, resuming an existing session fails with API 400 errors:

messages.X.content.0.thinking: each thinking block must contain non-whitespace thinking
Invalid 'signature' in 'thinking' block
Invalid 'data' in 'redacted_thinking' block

Root cause: The API validates signatures on thinking/redacted_thinking blocks, but these become invalid after migration. There's no graceful fallback — the entire session is unresumable.
Workaround: Manually strip all thinking and redacted_thinking blocks from the session JSONL file.
Expected behavior: Claude Code should handle this gracefully — either by stripping invalid thinking blocks automatically when replaying a session, or by not sending them to the API at all (since they're not needed for conversation continuity).
Environment: Claude Code on Linux, session migrated between machines (copied .claude/ and .claude.json*) - maybe have some official docs about how to do it properly ?

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

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.

Navigation Menu

Repository navigation

Chat history loss after PC restart, /resume recovers only 20-40% of context #23800

Description

Timeline

Steps to Reproduce

Expected Behavior

Actual Behavior

Environment

Additional data: `sessions-index.json` stops being updated

Observation

Facts

Additional finding: manually adding entries doesn't fix `/resume`

Possible investigation leads

Diagnostic scripts

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Chat history loss after PC restart, /resume recovers only 20-40% of context #23800

Description

Description

Timeline

Steps to Reproduce

Expected Behavior

Actual Behavior

Environment

Activity

Darkwing4 commented on Feb 6, 2026

github-actions commented on Feb 6, 2026

Darkwing4 commented on Feb 6, 2026

2jfs904judsw20600jikn613d0dookl23jsig commented on Feb 7, 2026

fredpottier commented on Feb 16, 2026

Additional data: sessions-index.json stops being updated

Observation

Facts

Additional finding: manually adding entries doesn't fix /resume

Possible investigation leads

Diagnostic scripts

Darkwing4 commented on Feb 16, 2026

Lennie commented on Feb 25, 2026

github-actions commented on Mar 25, 2026

github-actions commented on Apr 2, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Additional data: `sessions-index.json` stops being updated

Additional finding: manually adding entries doesn't fix `/resume`