Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableX: @RealMarvelX→ original

10 Open-Source Projects That Teach Claude Code to Remember Sessions and Count Tokens

Claude Code doesn't remember yesterday's sessions, and it doesn't show how much tokens cost. Tokens are the chunks of text you pay for. An X user has compiled 10 GitHub repositories that add memory, a cost tracker, and control over a real browser. The most popular, Anthropic's skills, has 192.5K stars. Start with token tracking and memory, and check the licenses of two of the projects before any commercial launch.

AI-processed from X: @RealMarvelX; edited by Hamidun News
10 Open-Source Projects That Teach Claude Code to Remember Sessions and Count Tokens
Source: X: @RealMarvelX. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @RealMarvelX ↗
Editor-in-Chief
Key metric #1ccusage counts tokens and costs from local logs
Key metric #2deja-vu indexes the records of 43 agents and passes sessions between them
Key metric #3Quest is under PolyForm Noncommercial, opcode under AGPL-3.0: review required

Claude Code writes code well, but it has everyday problems. It doesn't remember what you did yesterday, so every new launch starts with a recap of the project. It doesn't show how many tokens a task consumed, and tokens are the chunks of text you pay for.

And it can't open a browser on its own to check that a page works. The author of the roundup on X (@RealMarvelX) has collected 10 open-source repositories on GitHub that close these gaps. The most popular, Anthropic's skills, has collected 192.

5K stars: that's how GitHub users mark projects they like. The author describes the chain as: learn, remember, monitor, run, install. This is a social media post, not a verified review, but the set is practical.

Let's go through it by group.

Learning and standardizing: three repositories

skills (anthropics/skills) defines a single way for Claude Code to find and use skills: ready-made sets of instructions for a specific task, such as "how to format a report" or "how to work with spreadsheets." This is the most prominent project on the list, with 192.5K stars.

claude-code-best-practice (shanraisshan) is a map of Claude Code's capabilities: features, subagents (helpers for individual subtasks), and hooks (scripts that fire on specific events). It's useful as a cheat sheet if you only use part of what the tool offers.

Claude-Code-Quest (wesammustafa) teaches working with Claude Code through three levels of exercises of increasing difficulty. It suits onboarding new team members. Its license is PolyForm Noncommercial, which means commercial use is restricted.

Memory, accounting, and cost control: four repositories

claude-mem (thedotmack) carries useful context from one session into the next. Without it, you explain the project structure and past decisions to the agent again.

deja-vu (vshulcz) indexes session records from 43 different agents and helps pass sessions between them. If you switch between several AI assistants, you can find what was done last time and in which tool.

ccusage calculates token consumption and cost from local logs. For teams that pay for usage, it's the fastest way to see which tasks cost the most.

ccstatusline (sirmalloc) shows the model, the Git branch, and real-time token usage in the status line. It's not a after-the-fact report: the numbers are visible while you work.

Running, installing, and licenses

agent-browser (vercel-labs) gives agents control over a real browser. An agent can open a page, walk through a scenario, and verify the result instead of guessing from the code.

opcode manages Claude Code sessions and agents from a single interface. It's for people who run several tasks in parallel and don't want to lose track of them across different terminals.

A second skills (vercel-labs/skills) installs skills into Claude Code and other AI agents. The name matches Anthropic's repository, but the task is different: there it's the standard, here it's installation. Don't confuse them when searching.

The author separately warns about licenses. The PolyForm Noncommercial license on Claude-Code-Quest prohibits commercial use. The AGPL-3.0 license on opcode is a strong copyleft license: if you embed such code in a network product, you'll have to publish the source of your changes. For an internal tool the risk is lower, but for a product you sell to customers, you need a lawyer's consultation.

What to do in practice

The post contains no data on how stable each tool is or how it works alongside the others. GitHub stars indicate popularity, not quality. Connect the tools one at a time, and test them on a sandbox project first.

A sensible order is this. First install ccusage and ccstatusline to see where tokens go. Then add claude-mem so you don't have to recap the project every time. Only after that connect agent-browser and the shared skills: they extend capabilities, but your token consumption and the attack surface for errors or exploits grow too.

Story

Key Questions Answered

What are the key details on Learning and standardizing: three repositories?

skills (anthropics/skills) defines a single way for Claude Code to find and use skills: ready-made sets of instructions for a specific task, such as "how to format a report" or "how to work with spreadsheets." This is the most prominent project on the list, with 192.5K stars. claude-code-best-practice (shanraisshan) is a map of Claude Code's capabilities: features, subagents (helpers for individual subtasks), and hooks (scripts that fire on specific events).

What are the key details on Memory, accounting, and cost control: four repositories?

claude-mem (thedotmack) carries useful context from one session into the next. Without it, you explain the project structure and past decisions to the agent again. deja-vu (vshulcz) indexes session records from 43 different agents and helps pass sessions between them. If you switch between several AI assistants, you can find what was done last time and in which tool. ccusage calculates token consumption and cost from local logs. For teams that pay for usage, it's the fastest way to see which tasks cost the most.

What are the key details on Running, installing, and licenses?

agent-browser (vercel-labs) gives agents control over a real browser. An agent can open a page, walk through a scenario, and verify the result instead of guessing from the code. opcode manages Claude Code sessions and agents from a single interface. It's for people who run several tasks in parallel and don't want to lose track of them across different terminals. A second skills (vercel-labs/skills) installs skills into Claude Code and other AI agents.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…