DevDenThis week's trending AI repos are built for agents, and almost all come from companies
Nine of the ten fastest-rising AI repos on GitHub this week are built for AI agents to use, or for people to supervise agents. Nine of the ten also sit under a company or organization account. Only one comes from an individual developer.
The ten hold about 535,000 stars between them, per the November 14, 2026 list from the AI Employee Directory, a Japanese tracker. The English translation here is ours. Below, each repo gets what it does, who it suits, and the tracker's notes, with the repo link.
1. anthropics/skills (194K stars, Python)
https://github.com/anthropics/skills
The official collection of Agent Skills samples from Anthropic. A skill bundles instructions, scripts and reference material so Claude can repeat a task, such as building a document, the same way each time. The repo holds the skill specification, templates and examples, plus reference implementations of skills behind production document features such as PDF and Excel.
Suits: anyone who wants to teach Claude or Claude Code a document task or a routine procedure.
2. cloudflare/security-audit-skill (72K stars, MIT license)
https://github.com/cloudflare/security-audit-skill
Turns a coding agent into a tiered security auditor. The agent audits step by step, each finding is verified separately, and the results come out in a machine-readable format. To start, load the skill into your agent and run the audit flow against the target repository. The tracker's note on how much human review is needed reads unclearly in translation, so check the README.
Suits: development teams who want to check AI-written code for vulnerabilities before release.
3. ZhuLinsen/daily_stock_analysis (72K stars, Python)
https://github.com/ZhuLinsen/daily_stock_analysis
An LLM-based stock analysis system. It reads several market data and news sources, builds a screen to support decisions, and sends notifications automatically. It is the only repo on the list aimed at an everyday person with one concrete goal.
Suits: individual investors who want to automate their daily market check.
4. pingdotgg/t3code (60K stars, TypeScript)
https://github.com/pingdotgg/t3code
One place to control several coding agents, such as Claude Code, from both a computer and a phone.
Suits: developers who want to hand tasks to agents while they are away from their desk.
5. vercel-labs/agent-browser (51K stars, Rust)
https://github.com/vercel-labs/agent-browser
A command-line tool that lets an agent drive a browser: open pages, click and type text, all through commands.
Suits: developers who want to hand web tasks to an agent.
6. meta/astryx (27K stars, TypeScript)
https://github.com/meta/astryx
An open-source design system you can customize in detail, adapted so agents can work with it. The tracker links it to a use case where an app's design system is exported and used to create motion images of the interface with Remotion.
Suits: front-end developers who want AI-built screens to stay consistent with their design.
7. vercel-labs/agent-desktop (23K stars, Rust)
https://github.com/vercel-labs/agent-desktop
A command-line tool that lets an agent look at a computer screen and operate it. It turns "computer use", where an agent views and controls the screen, into something you can call from the command line.
Suits: developers building their own agents to automate screen tasks.
8. google/elephant-goldfish-model (19K stars, Python)
https://github.com/google/elephant-goldfish-model
An open-source plugin for designing software with agents and reviewing what they design. The tracker's heading for this repo ("representative-goldfish-model") doesn't match its GitHub address, so use the address above.
Suits: development teams who want agents involved from the design stage.
9. NVIDIA/auto-ontology (13K stars, Python)
https://github.com/NVIDIA/auto-ontology
A Python framework that lays a meaning-based graph over scattered data, so LLMs and agents can trace the reasons behind an answer. It generates fixed queries, pulls out context, and adds explainable reasoning to LLM and agent workflows.
Suits: people who want explainable search and reasoning in an agent workflow.
10. GoogleChrome/modern-web-guidance-src (4.9K stars, TypeScript)
https://github.com/GoogleChrome/modern-web-guidance-src
Keeps teaching coding agents the latest web best practices.
Suits: teams who want to protect the quality of front-end code that agents write.
What the list says
The user of open source is changing. These repos are skills, command-line tools and guidance files meant to be loaded into an agent, not libraries a developer imports. agent-browser and agent-desktop are tools whose main user is a model.
Supervising agents is now its own category. t3code manages many agents at once. The security audit skill and the design review plugin check what agents produce. The question has moved from "can an agent do this?" to "how do I keep track of what the agents did?"
Skills are the format to watch. Anthropic's skills repo has about 2.7 times the stars of the next repo, and Cloudflare's audit repo uses the same format. Companies appear to be adopting it, not only Anthropic.
Companies lead. Meta, Cloudflare, Anthropic, Google, Vercel Labs, NVIDIA, GoogleChrome and Ping own nine of the ten. The one personal project, the stock analyzer, is also the only repo built for a person and not an agent.
What we haven't checked
We did not install or run any of these. Star counts and descriptions come from the tracker, which says the counts were taken from GitHub on November 14 and the descriptions were summarized from each README. Its t3code entry shows two slightly different counts (60K in the list, 62K on the detail entry); we used 60K. Read each README and license before you use anything. Only the security audit repo has its license (MIT) stated by the tracker.
Source
Based on the AI Employee Directory's GitHub Trending page, updated November 14, 2026: https://meikan.moveai.jp/trending/?from=x_ainsight. The directory also shares the list on X as @so_ainsight. The summaries are rewritten, and the analysis is ours. English translation by DevDen.