Six Claude Code Terminals, One Crash, and Hours of Lost Work

Saving context to a CLAUDE.md file or to memory does not protect a Claude Code session from a crash, and this video is a 39-second case study in learning that the hard way. Kathryn (Camply) runs six Claude Code terminals at once across different projects. Her computer crashed, and a project she had been building for days came back blank. When she reopened the chat, Claude Code told her it did not know what they had been talking about. She closes by asking her audience directly: how do you work across multiple terminals and keep context so a crash does not wipe everything?

The Workflow That Failed

Her setup is common among founders building with AI tools: several terminal tabs open simultaneously, each running its own Claude Code session on a different project. One of those sessions had been running for days. The video is a straight-to-camera selfie with caption overlays, no screen recording, which fits the format: this is a help request, not a tutorial. The on-screen text opens with "I just lost hours of work in Claude Code because I had 6 terminals open and my computer crashed."

What She Already Tried

She did the two things most people reach for. She asked Claude to save the project state to a context file, and she asked it to save to memory. Neither held. After the crash, reopening the terminal produced a session with no awareness of the prior conversation. That result points to the gap in her workflow: context files and memory store facts and preferences, but they do not store the conversation itself. A fresh session reads those files, but it does not automatically become the old session.

The Fix She Is Asking For Exists

Claude Code saves every session to local disk automatically, per directory. The recovery path after a crash is claude --continue (or -c), which reloads the most recent conversation in the current directory with full message history and tool results, or claude --resume (or -r), which opens a picker to select any prior session. From inside a running session, /resume switches to a different conversation. For a six-terminal setup, the discipline that makes this work is running each project in its own directory, so each terminal's session history stays separate and resumable on its own.

Why the Memory File Approach Still Matters

Her instinct was not wrong, just incomplete. A CLAUDE.md file is the right place for durable project context: architecture decisions, conventions, current state. It makes any new session smarter from the first message. But it is a summary layer, not a transcript. The robust version of her workflow uses both: CLAUDE.md for standing context that survives everything, and --resume for recovering the actual conversation thread after a crash or accidental close.

Key Takeaways

• Asking Claude Code to "save to memory" or write a context file does not preserve the conversation itself; those files inform new sessions but do not restore old ones.
• Claude Code automatically saves session history locally per directory. claude --continue reloads the latest session, and claude --resume lets you pick from prior sessions.
• Running each project in its own directory keeps multi-terminal session histories separate and individually recoverable.
• Use CLAUDE.md for durable project state and session resume for conversation recovery. They solve different problems.
• The video drew its value from the comments format: state the failure specifically, ask what you are doing wrong, and let the audience supply the fix.

Resources

• Claude Code docs: Manage sessions covers --continue, --resume, and what gets restored.