Here is today’s look at what developers are starring on GitHub, plus the latest product updates from the GitHub Changelog. Two themes stand out on 21 November 2026: spec-driven development for AI coding assistants is pulling in big numbers, and agent skills collections keep climbing. We’ve also picked a self-hosted project tracker and a long-running PHP asset manager for some variety.
Star counts below are as shown on GitHub Trending and the daily language pages (Python, JavaScript, TypeScript, PHP) on 21 November 2026. They change quickly, so treat them as a snapshot. Repos featured in our last few roundups are skipped.
Trending repositories
1. github/spec-kit
What it does: GitHub’s toolkit “to help you get started with SDD or any other process”. SDD here is spec-driven development: you write a specification first and let a coding agent plan and implement from it.
- Language: Python
- Stars: 150,908 (127 today)
Why it matters: “vibe coding” falls apart on larger features. Spec-first workflows give agents a stable reference to work from and give humans something reviewable before any code is written. Having one of these come from GitHub itself makes it a useful baseline to compare others against.
2. Fission-AI/OpenSpec
What it does: “Spec-driven development (SDD) for AI coding assistants.” It’s another take on keeping agents aligned with a written spec.
- Language: TypeScript
- Stars: 75,415 (87 today)
Why it matters: with spec-kit and OpenSpec both trending, it’s worth trying each on a small feature in your own repo. The approach that fits your team’s review process is the one that will actually stick.
3. Imbad0202/academic-research-skills
What it does: academic research skills for Claude Code, covering a research → write → review → revise → finalise pipeline.
- Language: Python
- Stars: 57,968 (100 today)
Why it matters: it shows how far the “skills” format now stretches beyond coding. If you write technical docs, papers or long reports, a staged pipeline like this is a good template for your own agent workflows.
4. vercel-labs/agent-skills
What it does: “Vercel’s official collection of agent skills.”
- Language: JavaScript
- Stars: 34,649 (71 today)
Why it matters: vendor-maintained skills are a good way to teach a coding agent a platform’s conventions without stuffing everything into one long prompt. If you deploy on Vercel or use Next.js, browse these before writing your own.
5. phone-harness/phone-harness
What it does: “let your agent control your phone.” It’s a harness for driving a phone from an AI agent.
- Language: Python
- Stars: 5,422 (117 today)
Why it matters: mobile is still a gap for most agent tooling, which is largely browser- and terminal-based. Projects like this are interesting for mobile QA automation and accessibility experiments. As always with agents that control real devices, test on a spare phone first.
6. SWE-agent/mini-swe-agent
What it does: “The 100 line AI agent that solves GitHub issues or helps you in your command line.” Its description promises no huge configs and no giant monorepo, and claims a score above 74% on SWE-bench.
- Language: Python
- Stars: 9,924 (37 today)
Why it matters: if you want to understand how coding agents work, a small, readable agent is a much better teaching tool than a large framework. It’s also a handy baseline for comparing your own agent setup.
7. croffasia/itsaplan
What it does: an open-source, self-hosted alternative to Linear and Plane for project management and issue tracking “where teams and AI agents work side by side”.
- Language: TypeScript
- Stars: 2,467 (29 today)
Why it matters: teams that need their issue tracker on their own infrastructure, for compliance or cost reasons, have relatively few modern options. This one is designed with agents as first-class participants.
8. grokability/snipe-it
What it does: “A free open source IT asset/license management system.”
- Language: PHP
- Stars: 15,218 (8 today)
Why it matters: not every trending repo is new. Snipe-IT is a mature, self-hosted way to track laptops, licences and accessories, and a good real-world PHP codebase to learn from if you work with Laravel.
GitHub product updates
From the GitHub Changelog, posted on 20 November 2026:
Security advisory permissions are now generally available
Organisation owners can now let people work on repository security advisories, including private vulnerability reports, through custom repository roles, without making them repository admins or organisation-wide security managers. Four new permissions are available under Security: view advisories; create, edit and manage advisories; accept private vulnerability reports and close or reopen advisories; and publish advisories and request CVEs. Merging from a temporary private fork still needs a repository admin or security manager. This is available on plans that support custom repository roles, including GitHub Enterprise Cloud.
GitHub Enterprise Server 3.23 release candidate
Highlights include OR, explicit AND and parentheses in GitHub Projects filters; AI secret detection running fully on the appliance (public preview, including air-gapped setups); malware advisories for PyPI, Maven, RubyGems, NuGet, Go, Rust and Composer in the Advisory Database; a default three-day Dependabot cooldown before version update PRs; expiring OAuth app tokens (public preview); and the option to move the Redis primary to a dedicated node.
Pin your favorite repositories to your dashboard
The new dashboard has a Favorite repositories section. Pick repos with Edit or Add repositories; favourites sync with pinned repositories in GitHub Mobile, up to 25 across both.
Takeaway
Spec-driven development is having a moment: GitHub’s spec-kit and OpenSpec are both trending, so if your agent-written code keeps drifting from what you meant, try writing the spec first. On the platform side, the new security advisory permissions are worth setting up if you maintain popular open-source repos, because they let you bring in trusted triagers without handing out admin rights.
