Context persistence across sessions - major workflow disruption for complex projects

a nice feature would be able to save context checkpoints, which can be invoked with a slash command.

This way you can build context relevant for a specific task, save the checkpoint, and then resuse it at will.

Currently the way i do this is by building context BEFORE describing any task - this includes allowing claude to do some relevant research (i.e explain how X system integrates with Y, and report back to me with a detailed guide to the data structures and function signatures") then, I will provide a task. After the task is complete, I use esc to revert/fork the conversation back to the initial task assignment - which I then replace with the given task.

This works great, but is clunky given the circumstances - I can only run one task at a time, unless i use the --resume feature to reopen the conversation in a new session. However, the --resume selection interface is clunky and not very easy to differentiate between multiple conversation with the same initial context (5 word summaries are far from helpful lol)

After writing this out, I realise something similar could be achieved by asking claude to export his understanding to a local slash commands, which may be a viable option now that the support file references and bash scripts..

This is exactly the pattern I've been investigating! You've discovered what I call the "context checkpoint" problem - AI agents work best with fresh context but need persistent knowledge.
Your manual workflow (build context → save checkpoint → fork for tasks → ESC to revert) is brilliant but painful. I've been building something that automates this exact pattern.
The key insight: AI agents naturally create better specifications when they know they're handing off to a "fresh" instance. Your approach of building context BEFORE describing tasks forces this beneficial behavior.
I'm currently running a private beta (25 spots) for a tool that handles this automatically - preserves context across sessions while keeping each task execution fresh. Would love your feedback since you clearly understand the problem space: [discord.gg/k6kkPS72](url)
The "export to local slash commands" idea is interesting but still requires manual orchestration. What we really need is automatic context preservation with easy branching/forking, which sounds like exactly what you're asking for.

a nice feature would be able to save context checkpoints, which can be invoked with a slash command. This way you can build context relevant for a specific task, save the checkpoint, and then resuse it at will. Currently the way i do this is by building context BEFORE describing any task - this includes allowing claude to do some relevant research (i.e explain how X system integrates with Y, and report back to me with a detailed guide to the data structures and function signatures") then, I will provide a task. After the task is complete, I use esc to revert/fork the conversation back to the initial task assignment - which I then replace with the given task. This works great, but is clunky given the circumstances - I can only run one task at a time, unless i use the --resume feature to reopen the conversation in a new session. However, the --resume selection interface is clunky and not very easy to differentiate between multiple conversation with the same initial context (5 word summaries are far from helpful lol) After writing this out, I realise something similar could be achieved by asking claude to export his understanding to a local slash commands, which may be a viable option now that the support file references and bash scripts..

Your checkpoint idea is exactly what's needed! I've been working on a related solution that currently provides isolated persistent sessions per project: https://github.com/stefans71/Persistent_Claude

Your checkpoint workflow is brilliant - I'm adding it to the roadmap. The idea of saving context states BEFORE tasks and reusing them would solve so many workflow issues.

Currently my tool just maintains one persistent session per project, but adding checkpoint save/load functionality would be the logical next step. The export to local files idea is also great for team collaboration.

Would love your thoughts on the current implementation and what checkpoint features would be most valuable!

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

We built an open-source workaround using the hooks API: claude-memory-bridge

How it works

• PreCompact hook: Saves session context (initial request + key decisions) to SQLite before compaction
• SessionStart hook: Reinjects last 5 sessions as "Previous Session Context"
• Project-scoped: Memory is keyed by working directory, so projects don't leak into each other

Quick start

git clone https://github.com/Equilateral-AI/claude-memory-bridge.git
cd claude-memory-bridge
npm install && npm run install-hooks

It's a basic workaround - not a replacement for proper native support. Would love to see this built into Claude Code officially.

The hooks API makes this possible, but automatic summarization and smarter context management would be much better as a first-class feature.

I built an open-source solution for this: https://github.com/amarodeabreu/claude-graph-memory.

It uses a local Neo4j-compatible graph to persist project context. Happy to get feedback.

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue was closed incorrectly despite recent human comments. This behavior of the bot is reported at https://github.com/anthropics/claude-code/issues/16497. Please upvote that issue, so maybe it gets noticed.

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.