Here is today’s look at what developers are starring on GitHub and what changed on the platform itself. Repository data comes from GitHub Trending (daily view, plus the Python, JavaScript, TypeScript and PHP language pages), captured on the evening of 11 November 2026 (IST). Star counts are exactly as shown on the trending page at that time and will have moved since. We skipped repositories featured in our recent roundups.

Trending repositories

1. browser-use/browser-use

What it does: An open-source Python library for building AI agents that control a real web browser, with an optional hosted cloud offering for stealth browsers and an agent API.

Language: Python · Stars: 120,706 (171 today)

Why devs care: “Let the agent use the website” is becoming a standard automation pattern for sites without APIs. This is one of the most widely used libraries for doing that from Python.

2. shadcn/improve

What it does: An agent skill that audits a codebase with your most capable model and writes self-contained implementation plans for cheaper models (or humans) to execute. It never implements anything itself; “the plan is the product”. Install with npx skills add shadcn/improve.

Language: not listed (Markdown skill) · Stars: 10,570 (333 today)

Why devs care: It is a practical take on cost control with AI coding agents: spend expensive tokens on judgement and specs, cheap tokens on execution. Commands such as /improve security and /improve branch make it easy to try on a real repo.

3. GoogleChrome/modern-web-guidance-src

What it does: The source repository for Modern Web Guidance, a set of agent skills that embed web platform best practices and browser-compatibility data into coding agents, steering them away from legacy patterns. It includes the tooling for authoring, calibrating and evaluating that guidance.

Language: TypeScript · Stars: 1,530 (133 today)

Why devs care: It is stewarded by the Chrome, Edge, Firefox and MDN teams together. If your AI assistant keeps suggesting outdated JavaScript or CSS, this is the cross-browser effort aimed at fixing that.

4. we-promise/sure

What it does: A self-hostable personal finance app (“for everyone, by everyone”) with dashboards for cash flow, net worth and spending, plus an AI assistant for questions about your finances.

Language: Ruby · Stars: 12,649 (1,079 today)

Why devs care: It was the biggest gainer in today’s overall list among projects we had not featured. It is also a sizeable, real-world Rails codebase (AGPL-3.0) worth reading if you build data-heavy web apps.

5. backnotprop/plannotator

What it does: A workspace for staying in the loop with coding agents: annotate plans, specs, Markdown and HTML artifacts, review diffs and PRs, share with your team, and send all feedback back to the agent in one click. It works with tools such as Claude Code, Codex, Copilot CLI, Gemini CLI and OpenCode.

Language: TypeScript · Stars: 10,039 (51 today)

Why devs care: As agents produce more code and plans, review becomes the bottleneck. A structured place to give feedback beats pasting comments into a chat box.

6. experientiallabs/experiential

What it does: An open-source gateway and router for agent workflows. It exposes hosted, bring-your-own-key and local models through one OpenAI-compatible API, controls which users and agents can use which models and how much they can spend, and can turn production traffic into a custom router or model.

Language: Python · Stars: 16,516 (675 today)

Why devs care: Per-user budgets and model access rules are exactly what teams need once more than a handful of people use LLM APIs. Getting started is pip install experiential followed by exp.

7. tiagozip/cap

What it does: A privacy-first, self-hosted CAPTCHA alternative. Instead of image puzzles, the visitor’s browser quietly solves small proof-of-work and instrumentation challenges in the background.

Language: JavaScript · Stars: 8,333 (32 today)

Why devs care: No third-party tracking, no “click the traffic lights”, and it can be configured to slow down autonomous AI agents that burn your resources. It runs with Docker or on serverless workers.

8. kunchenguid/lavish-axi

What it does: An editor for HTML artifacts, built on the idea that “HTML is the new markdown” when a rich document needs more than Markdown can express. It is published on npm and supports macOS, Linux and Windows.

Language: JavaScript · Stars: 4,534 (50 today)

Why devs care: AI tools increasingly produce HTML reports and artifacts. A dedicated editor for polishing them is a niche that is clearly growing.

Today’s themes

  • Agent skills everywhere. Two of today’s picks (improve and Modern Web Guidance) are skills for coding agents, following the Agent Skills format.
  • Keeping humans in control. Plannotator and Experiential are about review, budgets and governance rather than raw model power.
  • Self-hosting is alive. Sure and Cap both lean on “run it yourself” as a feature, not an afterthought.

GitHub changelog: the last 48 hours

  • Gemini 4 Carbon in GitHub Copilot (10 Nov) – Google’s new model is generally available for Copilot Pro+, Max, Business and Enterprise users across VS Code, Visual Studio, the Copilot CLI, the coding agent, JetBrains IDEs, Xcode, Eclipse and more. Rollout is gradual, and Business/Enterprise admins control access through model policies.
  • Dependabot alerts will pause on inactive repositories (10 Nov) – From 8 December 2026, GitHub will turn off Dependabot alerts on most repositories with no activity in the past year. Alerts are archived, not deleted, and any push turns them back on. Repositories with an enforced security configuration that enables alerts are not affected.
  • A more informative GitHub dashboard (10 Nov) – Less padding, single-line titles, six pull requests and issues by default instead of three, the option to pin Feed or Home as your default tab, and the agent sessions section hides when you have none.
  • GitHub App support for Enterprise Copilot seat management (10 Nov) – Enterprise GitHub Apps can call the enterprise-level Copilot seat APIs using a new fine-grained permission, “Enterprise Copilot license management”.
  • C++ language server auto-configures MSBuild builds (10 Nov) – The Microsoft C++ Language Server for Copilot CLI now bundles and runs its MSBuild compile-commands extractor automatically, so agents get go-to-definition and references without an extra setup step.
  • Secret scanning detects secrets in GitHub Actions workflow logs (9 Nov) – Secret scanning now also looks at Actions workflow logs, a common place for tokens to leak by accident.

What to do this week

  • If you maintain older but still-used repositories, push a commit or enforce a security configuration before 8 December so Dependabot alerts stay on.
  • Copilot Business/Enterprise admins: decide whether Gemini 4 Carbon should be enabled under your model policy, since new models are enabled by default unless you turned that off.
  • Pick one trending repo above and try it on a side project. Today, shadcn/improve is an easy one to evaluate in under an hour.

Key takeaway

Today’s trending list is dominated by tools that make AI agents more useful and more controllable: browser automation, planning skills, review workspaces and model gateways. On the platform side, Gemini 4 Carbon arrived in Copilot and Dependabot will start pausing alerts on inactive repositories from 8 December. Check the GitHub changelog for the full list of updates.