[Resource]: Claude Session Restore

[ZENG3LD]

Display Name

Claude Session Restore

Category

Tooling

Sub-Category

General

Primary Link

https://github.com/ZENG3LD/claude-session-restore

Author Name

ZENG3LD

Author Link

https://github.com/ZENG3LD

License

Other (specify below)

Other License

MIT OR Apache-2.0

Description

Efficiently restore context from previous Claude Code sessions by analyzing session files and git history. Features multi-factor data collection across numerous Claude Code capacities with time-based filtering. Uses tail-based parsing for efficient handling of large session files up to 2GB. Includes both a CLI tool for manual analysis and a Claude Code skill for automatic session restoration.

Validate Claims

Install the CLI tool and run it on any existing Claude Code session file from ~/.claude/projects to verify multi-vector data extraction and context restoration capabilities.

Specific Task(s)

After installing the tool, list recent sessions to see extracted data from all event types (tasks, messages, tools, bash commands), then load a specific session to view complete context restoration including git branch information.

Specific Prompt(s)

Run: session-summary list --max-age-hours 24
Then select a session and run: session-summary load "<path-to-session.jsonl>"
This will demonstrate multi-vector data extraction and full context restoration.

Additional Comments

The tool provides two usage modes: a standalone CLI tool (session-summary) for manual session analysis, and a Claude Code skill that enables voice-activated session restoration with the "restore session" command. Supports multiple languages and includes efficient reverse parsing for large session files.

Recommendation Checklist

• I have checked that this resource hasn't already been submitted
• My resource provides genuine value to Claude Code users, and any risks are clearly stated
• All provided links are working and publicly accessible
• I am submitting only ONE resource in this issue
• I understand that low-quality or duplicate submissions may be rejected

[hesreallyhim]

thanks @ZENG3LD this looks very cool, much appreciated. I just wanted to ask real quick since I noticed the first few issues you opened were laking the appropriate labels and I'm really trying to understand whether there is a bug in the issue labeler/processor workflow or not, because this is happening loads of times (not just you) - did you use the gh cli to submit this resource? If you don't mind, could you tell me exactly how you opened those earlier issues? They look well-formed and everything, so it would be helpful to make sure my bot is not acting up. It would be helpful to know that's all. 👍🏽

[ZENG3LD]

Hi @hesreallyhim! Thanks for reviewing the submission.

Regarding the earlier issues (#555, #556) - yes, I used gh CLI (gh issue create and gh api). Both were immediately flagged with needs-template and auto-closed.

The issue is that the resource-submission label is only applied automatically when creating an issue through the web form template. When using CLI, even with properly formatted markdown body, the label doesn't get added, so your validation workflow doesn't recognize it as a valid submission.

Your detect-informal workflow then kicks in, sees an issue without resource-submission that looks like a resource recommendation, and applies needs-template + closes it.

For #557 (current), I used the web form directly, which auto-applied the resource-submission label and passed validation.

TL;DR: The bot requires issues to have the resource-submission label, which only gets applied via the web form interface, not through CLI submission.

Hope this helps identify the issue! 👍

[hesreallyhim]

Thanks for the clarification @ZENG3LD that's what I suspected I guess I didn't realize so many people are using GH CLI to open issues. Good to know.

EDIT: unless GH CLI actually presents the form template to you, my current thinking is to not support it fwiw.

[hesreallyhim]

@ZENG3LD wow interesting sounds very powerful - maybe a little too powerful (I wish Claude would forget its dumb behavior sometimes) - i think this might be unique somewhat to your workflow (which is fine, it doesn't mean it's an uncommon workflow) - but I just checked my largest convo file is about 40mb and you support up to 2GB(??) so i'm guessing you prefer to use continue/resume a lot(?)

Anyway, looks cool thank i will try to give it a spin

[ZENG3LD]

Yes, if you use a master session with an agent carousel (research, implementation, writing integration tests, debugging by tests until getting real data), they can significantly overflow context files. The problem is that when this happens, VSCode extensions for Claude break, and they simply can't be loaded. Had to make a small utility + skill for memory extraction.

Don't know how common this problem is, but in any case the idea was to make a lightweight tool that can access memory in the native .claude/projects/ storage without needing to fully parse it and put it into a .db, but at the same time a slightly more convenient and simple sequence of actions than explaining to Claude that it needs to find all recent sessions by modification date, look at N-loc, give a brief summary so I can decide which context to restore, after that it inevitably overflows its memory with special symbols and tags from the session file, and then you also need to explain what to look for in the git log. But here it's all with one command "restore session"

[hesreallyhim]

@ZENG3LD ok cool, I'm not sure about this "master session" pattern but maybe it's a thing. I think I'm going to give this a try. Can you confirm that the crypo addresses on your readme are personal wallet addresses and and that you are not promoting or involved in any token/airdrop/etc. schemes?

[ZENG3LD]

Hey, thanks for the question - totally fair to ask!

About the "master session" pattern:

This is my personal approach to orchestrating work with Claude Code. The idea is that the main Opus agent is essentially forbidden from doing anything directly - for any action, it spawns specialized Sonnet agents. These agents don't bloat its context memory because from its perspective they're just blocking tasks. This keeps the main agent in context longer without auto-compaction. The agents are created inside it, so all their work gets saved to the master session file (which is why it grows large).

The alternative is Ralph (an autonomous loop tool I also use and might switch to more in the future). As a side project, I'm building a wrapper for claude-code, codex, and gemini-cli that provides remote access via GUI, webapp, or Telegram mini-app. The natural evolution is to integrate Ralph and memory management directly into that. I initially tried parsing sessions into a .db but it was overkill, so I went with something lighter. Still figuring out the best approach for Ralph pipelines since it creates many small autonomous files rather than one bloated master session.

About the crypto donation addresses:

Yes, those are my personal wallet addresses. I'm from Russia where GitHub Sponsors isn't supported, and I haven't set up Patreon/Ko-fi/etc. yet - crypto was just the simplest option for now.

Not involved in any token/airdrop schemes:

I have zero connection to any crypto projects. I'm primarily developing a trading terminal (not crypto-specific - think something like OpenBB but full Rust, from texture rendering to execution). It uses crypto data streams for testing because they're free, but the terminal itself is broader than just crypto. It's partially open-source, partially source-available, monetized only through donations at this stage. If I ever add a subscription service, it would be me personally as the org owner - not any third-party crypto-whatever-scammers.

[github-actions]

🤖 Validation Results

✅ All validation checks passed!

Your submission is ready for review by a maintainer.

Validated Data:

{
  "display_name": "Claude Session Restore",
  "category": "Tooling",
  "subcategory": "General",
  "primary_link": "https://github.com/ZENG3LD/claude-session-restore",
  "author_name": "ZENG3LD",
  "author_link": "https://github.com/ZENG3LD",
  "license": "NOASSERTION",
  "description": "Efficiently restore context from previous Claude Code sessions by analyzing session files and git history. Features multi-factor data collection across numerous Claude Code capacities with time-based filtering. Uses tail-based parsing for efficient handling of large session files up to 2GB. Includes both a CLI tool for manual analysis and a Claude Code skill for automatic session restoration.",
  "secondary_link": "",
  "last_modified": "2026-01-26:07-54-17",
  "active": "TRUE",
  "last_checked": "2026-01-29:01-26-08"
}

---

_{This comment is automatically updated when you edit the issue.}

[hesreallyhim]

@ZENG3LD I made one small alteration to reduce length and because from what I could you're doing multi-"factor" analysis, not vector in the sense of vector embeddings - but maybe I'm wrong, I'm gonna merge this in and I can just correct it later if you prefer. I'd love to chat with you more about this if you're interested, feel free to email me. (Also I think you could use a catchier name)

/approve