Here is today’s look at what developers are starring on GitHub, taken from the live GitHub Trending page and the daily language pages for Python, JavaScript, TypeScript and PHP on 17 November 2026. AI coding agents still dominate the lists, but today they share space with security tooling, a tiny LLM you can train yourself, and a couple of long-standing open-source favourites. Star counts are the totals shown on the trending pages when we checked, so they will have grown by the time you read this.
Trending repositories
1. obra/superpowers
Language: Shell · Stars: 319,089 (1,905 today)
Superpowers describes itself as “a complete software development methodology for your coding agents”, built from composable skills plus starter instructions that make sure the agent uses them. Instead of writing code immediately, the agent first asks what you’re really trying to build, then shows you the spec in short chunks.
Why devs care: it was the most-starred repo on the overall daily list today, and its README has setup instructions for many agents and CLIs. If your AI pair programmer tends to rush into code, this is a structured way to slow it down.
2. stablyai/orca
Language: TypeScript · Stars: 127,553 (1,052 today)
Orca bills itself as “the AI Orchestrator for 100x builders”. It runs Claude Code, Codex and other coding agents in parallel, each in its own git worktree, in a chat UI or the raw terminal, with diffs, a browser and a CLI in one app. It can also run agents on SSH hosts or cloud VMs, and has a mobile app that pings you when an agent needs input.
Why devs care: running several agents at once without them trampling each other’s files is a real problem, and worktree isolation plus line-level review notes is a practical answer.
3. elementalsouls/Claude-BugHunter
Language: Python · Stars: 8,574 (1,007 today)
A Claude Code skill bundle for bug hunting and external red-team work. The README lists 83 skills, 15 slash commands and 681 disclosed-report patterns across 24 vulnerability classes, plus engagement-folder scaffolding and Burp MCP integration.
Why devs care: it is a big, organised library of web-security testing knowledge for an AI assistant. Use it only on systems you’re authorised to test, such as your own apps, bug-bounty scopes or training targets like DVWA and OWASP Juice Shop.
4. usestrix/strix
Language: Python · Stars: 76,260 (325 today)
Strix is an open-source AI penetration-testing tool. Its agents run your app dynamically, try to find vulnerabilities, and confirm them with working proofs-of-concept. The README highlights multi-agent orchestration, a developer-first CLI, auto-fix suggestions, and CI/CD integration for scanning pull requests.
Why devs care: validated findings with a proof-of-concept are much easier to act on than a long list of static-analysis warnings, which makes it a good fit for teams without a dedicated security tester.
5. jingyaogong/minimind
Language: Python · Stars: 65,763 (188 today)
MiniMind’s goal, in its own words, is to train a roughly 64M-parameter language model completely from scratch for about ¥3 of compute and around two hours of training. The code covers the whole pipeline: data cleaning, pretraining, SFT, LoRA, DPO, RLAIF methods, tool use and distillation.
Why devs care: it is one of the most approachable ways to see every stage of LLM training in readable code on a single consumer GPU, which is great for learning before you fine-tune larger models.
6. sveltejs/svelte
Language: JavaScript · Stars: 89,435 (533 today)
Svelte is a compiler that turns declarative components into efficient JavaScript that updates the DOM surgically. It topped the daily JavaScript list today.
Why devs care: Svelte remains a popular choice for fast, lightweight UIs with less boilerplate. If you’ve only used React or Angular, it’s worth a weekend project.
7. cursor/plugins
Language: TypeScript · Stars: 16,795 (196 today)
The official Cursor plugin specification and a set of first-party plugins. Each plugin is a standalone folder with a .cursor-plugin/plugin.json manifest. Examples include a teaching plugin, a continual-learning plugin that updates AGENTS.md, a deep branch-review plugin, and a plugin scaffolder.
Why devs care: if you use Cursor, these show the expected plugin structure, and you can copy one as a starting point for your team’s own workflows.
8. glpi-project/glpi
Language: PHP · Stars: 6,629 (25 today)
GLPI (“Gestionnaire Libre de Parc Informatique”) is free IT asset management and ITIL service-desk software. It tracks computers, peripherals and licences, and handles incidents, requests and problems.
Why devs care: it led today’s PHP trending list. It’s a mature, self-hostable alternative to commercial IT asset and helpdesk tools, and a good real-world PHP codebase to study.
GitHub changelog: the latest updates
The newest entries on the GitHub Changelog when we checked were dated 16 November 2026. These had not been covered in yesterday’s roundup:
- Dependabot alert dismissal requests now last one year. With delegated alert dismissal, requests stay open for one year instead of one week, so reviewers have time to respond. It is live on github.com and will ship in GitHub Enterprise Server 3.24.
- Control where GitHub Copilot customizations come from. Enterprise admins can use the
strictPluginOnlyCustomizationmanaged setting to require that skills, custom agents, hooks, MCP servers or executable Copilot extensions come from plugins or managed settings. Pair it withstrictKnownMarketplacesto restrict marketplace sources. - VS Code retires the Local harness in February 2027. VS Code 1.146 deprecates the older Local harness for Copilot agent sessions. Its replacement, the Copilot harness, has been the default for new sessions since VS Code 1.143 and doesn’t move your files to the cloud.
- GitHub Copilot models deprecated on November 16, 2026. Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra are deprecated across most Copilot experiences. Suggested alternatives are Claude Opus 5.5 and GPT-6.1 Sol or GPT-6 Luna. Update any workflows that pin these models.
- Budget limit increased to 40,000 budgets per account. GitHub Enterprise Cloud accounts can now create up to 40,000 budgets, up from 10,000, with nothing to turn on.
Takeaway
Agent tooling keeps leading GitHub: methodology (Superpowers), orchestration (Orca), plugins (Cursor) and AI-driven security testing (Claude-BugHunter, Strix). If you try the security tools, keep them pointed at systems you own or are authorised to test. On the platform side, enterprise admins should check the Copilot model deprecations and the new plugin-only customization controls this week.
