GitHub Trending 2026-11-18: Popular repos are turning agents into handoff-ready workflows
This issue covers 2026-11-11 to 2026-11-18. The fastest-growing top three were Astryx (+8,700), Modern Web Guidance (+7,015), and anthropics/skills (+4,847). The top new repo, deepseek-ai/chat-demo, reached 4,563 stars. Stars are attention signals, not endorsements.
TL;DR
This week’s repos turn “agents can do things” into parts that can be handed off, checked, and constrained. Design systems give people and agents a shared language, web guidance scans legacy patterns, browser tools expose real sessions, and code review separates rules from model suggestions. For most readers, the sensible first move is a small reversible task, not handing over every permission.
本週三個判讀
1. The competition is moving from answers to handoffs
Astryx, Modern Web Guidance, anthropics/skills, and addyosmani/agent-skills all address the same operational problem: how the next person or agent understands the current components, rules, and workflow. Clear inputs, outputs, and a smallest check matter more than another chat surface.
2. Browser and review automation make permissions part of the product
agent-browser and Chrome DevTools MCP let agents see real browsers, while open-code-review places model suggestions inside a merge workflow. Sessions, allowed domains, deterministic findings, and human decisions should remain visible instead of becoming one black box.
3. The non-AI signal is bringing complex life back under user control
Sure, Wireless Device Battery Monitor, lastbook, and freebie-radar point to self-hosted finance, hardware, offline knowledge, and filtering. Self-hosting is not zero cost: backup, sync, licensing, supply chain, and deletion duties return to the user.
Fastest Growing — 本週星星增量 Top 10
| # | Repo | Weekly gain | Total stars | Created | Language | License |
|---|---|---|---|---|---|---|
| 1 | meta/astryx | +8,700 | 30,198 | 2026-01-09 | TypeScript | MIT |
| 2 | GoogleChrome/modern-web-guidance-src | +7,015 | 8,477 | 2026-01-27 | TypeScript | Apache-2.0 |
| 3 | anthropics/skills | +4,847 | 196,784 | 2025-09-22 | Python | Not listed |
| 4 | we-promise/sure | +4,786 | 16,284 | 2025-07-23 | Ruby | AGPL-3.0 |
| 5 | addyosmani/agent-skills | +3,884 | 118,682 | 2026-02-15 | JavaScript | MIT |
| 6 | alibaba/open-code-review | +3,675 | 53,795 | 2026-05-18 | Go | Apache-2.0 |
| 7 | ZhuLinsen/daily_stock_analysis | +3,653 | 73,108 | 2026-01-10 | Python | MIT |
| 8 | vercel-labs/agent-browser | +3,429 | 52,399 | 2026-01-11 | Rust | Apache-2.0 |
| 9 | tradermonty/claude-trading-skills | +2,985 | 6,884 | 2025-10-19 | Python | MIT |
| 10 | ChromeDevTools/chrome-devtools-mcp | +2,948 | 58,057 | 2025-09-11 | TypeScript | Apache-2.0 |
Top New Repos — 本週新誕生 Top 10
| # | Repo | Total stars | Created | Language | License |
|---|---|---|---|---|---|
| 1 | deepseek-ai/chat-demo | 4,563 | 2026-11-12 | TypeScript | MIT |
| 2 | krisgfx9/Wireless-Device-Battery-Monitor | 2,821 | 2026-11-16 | C++ | GPL-3.0 |
| 3 | apex-compute/ANR | 2,601 | 2026-11-11 | C++ | Not listed |
| 4 | Noniv/abyssal-garden | 2,182 | 2026-11-12 | JavaScript | MIT |
| 5 | Immortal-Protocols/lastbook | 1,820 | 2026-11-12 | Python | MIT |
| 6 | gylive/google-ultra | 1,737 | 2026-11-17 | Python | MIT |
| 7 | duwqijlk/muse-protocol | 1,596 | 2026-11-14 | Python | NOASSERTION |
| 8 | Eynnzerr/TelemetryCleaner | 1,504 | 2026-11-11 | Kotlin | MIT |
| 9 | erkamyaman/rust-anything | 1,477 | 2026-11-16 | Rust | MIT |
| 10 | ZiChuanLan/freebie-radar | 1,442 | 2026-11-12 | TypeScript | MIT |
值得深挖的專案
Astryx:A design system preparing an interface for agents
Astryx combines React components, themes, a CLI, documentation, and composable source. Its README still labels the project beta. It is worth trying in an internal tool or sandbox, but the small HN thread does not prove large-team maturity.
Modern Web Guidance:Best practices as an executable workflow
The official v1.0 release describes use-case guides, offline search, legacy scanning, and an agent skill. That is more actionable than asking an agent to “use modern Web,” though guidance can still lag platform changes. Use it as a review aid, not an automatic merge gate.
agent-browser:The real cost is permission
The CLI gives agents browser access, but cookies, payment pages, personal data, and irreversible clicks are the real risks. Start with a test site, fixed domains, and a test account before connecting it to daily automation.
open-code-review:Keep deterministic rules beside the model
Alibaba’s repository combines deterministic pipelines with an LLM agent and lists rules such as XSS and SQL injection. Compare it offline against existing team PRs first; “battle-tested” in a README is not your own false-positive rate.
Sure:Self-hosted finance still carries data responsibility
Sure is a self-hosted personal finance tool, while recent releases still include alpha builds, debug builds, provider sync, and encryption changes. Test import, backup, and recovery with fake data before using real accounts.
社群怎麼看
The Modern Web Guidance HN discussion has the strongest signal this week: readers support moving agents away from old JavaScript habits, while the discussion also exposes concerns about update speed and correctness. The Astryx discussion is better read as directional interest than proof of adoption. Together, the signal is support for operational workflows, with maturity and permission boundaries still open.
編輯台選擇
現在可試
Run Modern Web Guidance on a small reversible project, and prototype one internal-tool page with Astryx.
先觀望
Keep agent-browser, open-code-review, and Sure in a sandbox until you have environment-specific reliability and maintenance evidence.
不建議直接採用
Do not turn daily_stock_analysis or trading skills into investment decisions, and do not let browser agents touch payments, primary accounts, or irreversible settings. Treat TelemetryCleaner as a high-permission experiment requiring a full device backup and a rollback path.
Was this article helpful?



