VS Code extension crashes on startup when most recent session file is large

Description

The Claude Code VS Code extension (v2.1.34, darwin-arm64) crashes on startup when the most recent session .jsonl file is large (~467MB). The extension host terminates unexpectedly, triggering VS Code's extension bisect dialog. The CLI works fine with the same installation.

Steps to Reproduce

1. Use Claude Code extensively in a single session until the session .jsonl file grows to several hundred MB (in this case, 467MB / 9,781 lines at ~/.claude/projects/<project>/<session-id>.jsonl)
2. Close VS Code
3. Reopen VS Code — the extension attempts to load the most recent session and crashes

Observed Behavior

• Extension host terminates unexpectedly on startup
• VS Code log shows: MCP server "claude-vscode" Failed to fetch tools: MCP error -32601: Method not found
• The command claude-vscode.editor.openLast is not found (extension commands never register because initialization fails)
• Restarting VS Code, reinstalling the extension, and disabling all other extensions do not help
• The CLI (claude --resume <session-id>) can eventually load the same session, but takes a very long time

Workaround

Renaming the large session file resolves the issue:

mv ~/.claude/projects/<project>/<session-id>.jsonl ~/.claude/projects/<project>/<session-id>.jsonl.bak

After renaming, the extension starts up normally.

Expected Behavior

The extension should handle large session files gracefully — either stream them, skip loading on startup, set a size limit, or warn the user — rather than crashing the extension host.

Environment

• Claude Code: v2.1.34
• Platform: darwin-arm64 (macOS, Apple Silicon)
• VS Code: latest
• Session file size: 467MB (9,781 lines)
• Other large sessions in the same project: 553MB, 376MB, 238MB

Additional Context

• The session had previously run out of context and was continued via /resume (summary-based continuation), which suggests long-lived sessions can grow unbounded
• The MCP -32601 error appears to be a side effect of the crash, not the root cause
• The /doctor command in CLI shows no issues
• No conflicting environment variables (ANTHROPIC_API_KEY, MCP_TOOL_TIMEOUT) are set