Here is today’s look at what developers are starring on GitHub, plus the latest product updates from the GitHub Changelog. Agent tooling still dominates the trending lists, but today’s picks also include a classic self-hosted backend and a long-running PHP project.
Star counts below are as shown on GitHub Trending and the daily language pages (Python, JavaScript, TypeScript, PHP) on 18 November 2026. They change quickly, so treat them as a snapshot.
Trending repositories
1. pingdotgg/t3code
What it does: “The best way to control Claude Code, Codex and more from any computer or phone”, according to its description. It is a front end for driving coding agents from wherever you are.
- Language: TypeScript
- Stars: 64,937 (1,264 today)
Why it matters: if you run long agent sessions, being able to check and steer them from your phone is a real quality-of-life gain. It was one of the fastest-growing repos on the TypeScript list today.
2. affaan-m/ECC
What it does: “The agent harness performance optimization system”, bundling skills, instincts, memory, security and research-first development for Claude Code, Codex, OpenCode and other agents.
- Language: JavaScript
- Stars: 306,241 (1,109 today)
Why it matters: with over 300,000 stars it is one of the biggest repos in the agent-tooling space. Worth browsing for ideas on structuring agent memory and safety rules, even if you do not adopt the whole system.
3. Leonxlnx/taste-skill
What it does: agent skills that give coding agents “taste”: art direction, reference images and a browser check, instead of generic AI-looking pages.
- Language: JavaScript
- Stars: 118,227 (822 today)
Why it matters: the classic complaint about AI-built UIs is that they all look the same. If you generate front ends with agents, this tackles that problem directly.
4. p-e-w/heretic
What it does: training-free language model modification. Its description lists removing censorship, suppressing “slop” and reducing sycophancy, among other uses.
- Language: Python
- Stars: 41,906 (620 today)
Why it matters: it changes model behaviour without a fine-tuning run, which is interesting for ML engineers who work with open-weight models locally. Check the model’s licence and your own policies before modifying and redistributing any model.
5. HKUDS/CLI-Anything
What it does: “CLI-Anything: Making ALL Software Agent-Native.” The project aims to give software a command-line interface that agents can drive, with a companion CLI-Hub.
- Language: Python
- Stars: 59,472 (384 today)
Why it matters: agents work best with tools they can call from a shell. If you build internal tools, this is a useful reference for making them agent-friendly.
6. colbymchenry/codegraph
What it does: a pre-indexed code knowledge graph that syncs automatically when code changes, for Claude Code, Codex, Gemini, Cursor, OpenCode, Copilot and other agents.
- Language: TypeScript
- Stars: 77,964 (129 today)
Why it matters: agents waste a lot of tokens re-reading files to understand a codebase. A ready-made graph of symbols and relations can make them faster and cheaper on large repos.
7. appwrite/appwrite
What it does: “The open-source cloud for agents & devs”, including auth, databases, storage, functions, messaging, hosting, realtime, WAF and more.
- Language: TypeScript
- Stars: 58,147 (239 today)
Why it matters: a self-hostable backend-as-a-service is a solid choice when you want Firebase-style features without vendor lock-in. It showed up on today’s TypeScript trending list.
8. RSS-Bridge/rss-bridge
What it does: “The RSS feed for websites missing it.” It generates feeds for sites that do not offer one.
- Language: PHP
- Stars: 9,369 (16 today)
Why it matters: a small, practical PHP project you can self-host in minutes. Handy for following sites in your feed reader, and a readable codebase if you want to learn how scrapers and feed generators work.
Quick summary
| Repo | Language | Stars | Today |
|---|---|---|---|
| pingdotgg/t3code | TypeScript | 64,937 | 1,264 |
| affaan-m/ECC | JavaScript | 306,241 | 1,109 |
| Leonxlnx/taste-skill | JavaScript | 118,227 | 822 |
| p-e-w/heretic | Python | 41,906 | 620 |
| HKUDS/CLI-Anything | Python | 59,472 | 384 |
| colbymchenry/codegraph | TypeScript | 77,964 | 129 |
| appwrite/appwrite | TypeScript | 58,147 | 239 |
| RSS-Bridge/rss-bridge | PHP | 9,369 | 16 |
GitHub product updates
The GitHub Changelog had a busy day around Microsoft Ignite, with updates posted on 17 November (US time). The highlights:
- GPT-6.2 Astra and GPT-6.2 Sol in Copilot: OpenAI’s new models are generally available in GitHub Copilot, aimed at agentic coding and code review. They are billed at provider list pricing under usage-based billing.
- MAI Code 2 and MAI Code 2 Flash: Microsoft AI’s new coding models arrive in Copilot. MAI Code 2 targets complex multi-file work and debugging; Flash targets quick edits and explanations.
- Copilot code review gets Adaptive effort and secret detection: Adaptive picks the review depth for each pull request based on complexity and risk. Reviews can also draw on related repositories and flag exposed secrets before they reach the main branch. These are in public preview.
- Copilot code review outcomes in Repository insights: a new page in the Insights tab shows how many Copilot review comments were addressed, including high-severity ones.
- GitHub Copilot Evals in public preview: create repeatable tests for your skills, agents and MCP tools, for example with the
/evalcommand, and re-run them to catch regressions after changes. - Copilot app updates: the Copilot app now supports Azure DevOps repositories (public preview) and GitHub Enterprise Server.
- Copilot in JetBrains: adds a cloud sandbox and remote access (both public preview, rolling out in phases), plus voice dictation and browser use controls.
- Copilot CLI with Microsoft Entra sign-in: public preview on Windows and macOS, with Linux support to follow.
- Copilot Fusion updates and Copilot activity in Microsoft Agent 365 round out the list for enterprise users.
For most teams, the code review changes and Copilot Evals are the ones to try first. Adaptive review saves you from picking review depth by hand, and evals give you a way to check that your custom agents and skills still behave after you change them.
Takeaway
Agent tooling keeps topping GitHub Trending: remote control for coding agents (t3code), agent harnesses (ECC), better-looking AI front ends (taste-skill), code graphs for agents (codegraph) and agent-ready CLIs (CLI-Anything). On the product side, GitHub is focused on making Copilot’s code review and custom agents measurable. Star the repos that fit your stack and check back tomorrow for the next roundup.
