GitHub Weekly Wins: Claude Code & Codex CLI Updates

> **TL;DR:** This week's GitHub Weekly Wins spotlights two terminal-based AI coding agents: Anthropic's Claude Code (135,954 stars), which now supports generating artifacts like docs and UI mockups directly from a session, and OpenAI's Codex CLI (95,415 stars), which added a Record & Replay feature that turns a screen-recorded task into a reusable automation skill. Both are open on GitHub and free to try today.

Key Takeaways

- Claude Code leads this week's roundup with 135,954 GitHub stars and a new ability to produce artifacts — documents, UI mockups — straight from a terminal session. - OpenAI's Codex CLI sits at 95,415 stars and now ships Record & Replay, which converts a recorded task into a repeatable skill. - Both tools run locally in the terminal (or IDE) and are built around natural-language, agentic workflows rather than simple autocomplete. - Claude Code is written in Python; Codex CLI is written in Rust and licensed Apache-2.0. - Neither project's public repo currently lists topics/tags, so discovery mostly happens through direct search and community roundups like this one.

This Week's Picks: Two Terminal Agents Leveling Up

Every week, [GitHub Weekly Wins](https://speka.info/github-weekly-wins/) rounds up open-source AI tooling that's actually shipping — not vaporware, not roadmap slides, just repos with commits and stars to show for it. This edition centers on a rivalry that's been defining agentic coding all year: Anthropic's Claude Code against OpenAI's Codex CLI. Both got feature updates worth knowing about, and both remain squarely terminal-first tools built for developers who'd rather describe a task in plain English than click through a GUI.

The details below come from a roundup video, ["New AI Drops: AI does your work, Claude in Slack, cheap AI & more"](https://www.youtube.com/watch?v=SMIuRWtbO9Y) by [@TheNextNewThingAI](https://www.youtube.com/watch?v=SMIuRWtbO9Y), cross-checked against each project's live GitHub data.

Anthropic/Claude Code

What it is

[Claude Code](https://github.com/anthropics/claude-code) is Anthropic's agentic coding tool that lives in your terminal. It reads and understands your codebase, then executes routine tasks, explains complex code, and handles git workflows — all through natural-language commands instead of a fixed command syntax. You can run it in a terminal, inside an IDE, or invoke it by tagging @claude directly on GitHub. The project is written primarily in Python and currently sits at 135,954 GitHub stars, making it the most-starred repo in this week's roundup by a wide margin.

Why it matters

The headline update this cycle is that Claude Code now supports creating artifacts — documents, UI mockups, and similar outputs — directly from a session, rather than requiring a separate app or export step. That closes a gap for developers who use Claude Code not just to write and fix code, but to produce the accompanying documentation, diagrams, or interface drafts a project needs. It's a small addition on paper, but it reflects a broader shift: Claude Code is increasingly positioned as a full workflow assistant, not just a code-completion layer bolted onto a terminal.

That expanding footprint hasn't been without friction. Anthropic's own [Claude Mythos preview was linked to a spike in CVE severity](https://speka.info/blog/claude-mythos-preview-linked-to-cve-severity-spike) reports earlier this year, and [Alibaba reportedly restricted Claude Code internally over backdoor concerns](https://speka.info/blog/alibaba-reportedly-bans-claude-code-over-backdoor-fears). Neither story changes what's in this week's changelog, but they're useful context for teams deciding how much autonomy to hand an agent that can now also generate design artifacts. Anthropic has separately been investing in the artifact/design side of its product line — see the recent [Claude design overhaul covering Slides, Decks, and a free alternative](https://speka.info/blog/claude-design-overhaul-slides-decks-free-alternative) — so this update lines up with where the company has been putting effort.

Who it's for

Developers already working inside a terminal-heavy workflow who want one tool to handle code changes, git operations, and now lightweight document or UI output without switching apps. Teams evaluating Claude Code for broader use should weigh the artifact feature against the governance questions raised above, especially in regulated or security-sensitive environments.

GitHub

[github.com/anthropics/claude-code](https://github.com/anthropics/claude-code)

OpenAI/Codex

What it is

[Codex CLI](https://github.com/openai/codex) is OpenAI's lightweight coding agent, designed to run locally on your own machine rather than in the cloud. It's aimed at the terminal by default, though OpenAI also offers IDE integrations for editors like VS Code, Cursor, and Windsurf, plus a separate desktop app and a cloud-based version called Codex Web at chatgpt.com/codex. The CLI itself is written in Rust, released under the Apache-2.0 license, and has reached 95,415 GitHub stars — putting it a clear second in this week's roundup, but still one of the fastest-growing agentic tools on GitHub.

Why it matters

The standout addition this week is Record & Replay: a feature that lets a developer screen-record themselves performing a task once, then turns that recording into a reusable, repeatable skill Codex can execute again on demand. That's a meaningfully different approach to automation than writing a script or a prompt template — it lowers the barrier for non-programmers on a team to capture a workflow and hand it to the agent, and it gives programmers a fast way to encode one-off fixes into repeatable macros. Paired with the CLI's local-first design, it reinforces OpenAI's pitch that Codex is meant to feel like a tool you own and run yourself, not a service you rent.

It's also a direct answer to the kind of agentic-workflow features Claude Code has been shipping, and it's worth watching whether Record & Replay becomes a template other coding agents copy the way artifact generation has spread across the category.

Who it's for

Developers who prefer a local-first, permissively licensed (Apache-2.0) agent and want to automate repetitive terminal tasks without hand-writing scripts. The Record & Replay feature specifically suits teams standardizing recurring maintenance tasks — dependency bumps, log triage, repetitive refactors — across a codebase.

GitHub

[github.com/openai/codex](https://github.com/openai/codex)

How to Choose Between Them This Week

If your team is already inside the Anthropic ecosystem and wants an agent that can also spit out supporting documents or UI drafts, Claude Code's new artifact support is the more immediately useful update. If you'd rather record a task once and replay it — especially for non-code, repetitive terminal work — Codex CLI's new feature is purpose-built for exactly that. Both are open-source, both are free to install, and neither requires you to commit exclusively; plenty of teams run both side by side depending on the task.

For the full backstory and more releases like these, check the [GitHub Weekly Wins hub](https://speka.info/github-weekly-wins/), where we track star growth and feature drops across the open-source AI coding space every week.

Frequently Asked Questions

Which has more GitHub stars, Claude Code or Codex CLI?

Claude Code leads with 135,954 stars compared to Codex CLI's 95,415, as of this roundup.

What new feature did Claude Code add?

Claude Code now supports creating artifacts, such as documents and UI mockups, directly from within a session.

What is Codex CLI's Record & Replay feature?

It lets you screen-record yourself performing a task once, then converts that recording into a reusable skill Codex can execute again automatically.

What programming languages are Claude Code and Codex CLI written in?

Claude Code is written in Python; Codex CLI is written in Rust and released under the Apache-2.0 license.

Can I use Codex outside the terminal?

Yes — OpenAI also offers Codex integrations for IDEs like VS Code, Cursor, and Windsurf, a desktop app, and a cloud-based version called Codex Web.

Sources & Attribution

- Inspired by / watch the full breakdown: [New AI Drops: AI does your work, Claude in Slack, cheap AI & more](https://www.youtube.com/watch?v=SMIuRWtbO9Y) (@TheNextNewThingAI) - https://github.com/anthropics/claude-code - https://github.com/openai/codex