GitHub Trending 2026-11-11: Agents Move into Workspaces, Browsers, and Security Checks
This report covers 2026-11-04 to 2026-11-11. T3 Code led fastest-growing repositories with 11,979 new stars, while Hookshot led new repositories with 6,652 total stars. Stars signal attention, not endorsement.
TL;DR
Developers are turning agent capability into practical components: resumable workspaces, reusable skills, verifiable security audits, browser and desktop control, and experiments closer to real hardware. The projects worth testing are not simply the flashiest demos. They are the ones that make permissions, recovery, and data flow understandable.
Three readings this week
1. The agent battleground is moving from answers to handoff
T3 Code puts threads, permissions, source control, remote access, and background work in one workspace. agent-desktop and OpenMausBot push computer use toward the desktop. The practical question is now whether someone else can understand what happened and continue safely. For Shareuhack's fleet, the useful lesson is an invocation and recovery receipt, not another chat window.
2. Skills and security audits are becoming composable delivery units
mattpocock/skills, anthropics/skills, and Cloudflare's security-audit-skill package methods for reuse. That speeds up workflows, but also makes side effects easier to miss. Before adoption, check what a skill reads and writes, whether its output can be re-verified, and how to roll back a failure.
3. Giving agents access to the outside world multiplies both value and risk
agent-browser, agent-desktop, and Scrapling reach into browsers, desktops, and web data. That is a meaningful step beyond text, but sessions, site rules, personal data, and proxies cannot be hidden behind one automation button. These tools belong in small-scope tests before production.
Fastest Growing — Top 10
| # | Repository | Weekly stars | Total stars | Created | Language | License |
|---|---|---|---|---|---|---|
| 1 | pingdotgg/t3code | +11,979 | 55,853 | 2026-02-08 | TypeScript | MIT |
| 2 | cloudflare/security-audit-skill | +10,967 | 69,414 | 2026-06-18 | JavaScript | MIT |
| 3 | mattpocock/skills | +9,872 | 321,010 | 2026-02-03 | Shell | MIT |
| 4 | NVIDIA/auto-ontology | +9,621 | 11,489 | 2026-04-27 | Python | Apache-2.0 |
| 5 | vercel-labs/agent-desktop | +9,502 | 21,451 | 2026-02-19 | Rust | Apache-2.0 |
| 6 | milind-soni/OpenMausBot | +8,951 | 20,705 | 2026-08-11 | TypeScript | Apache-2.0 |
| 7 | D4Vinci/Scrapling | +7,291 | 101,192 | 2024-10-13 | Python | BSD-3-Clause |
| 8 | anthropics/skills | +5,014 | 192,409 | 2025-09-22 | Python | Undeclared |
| 9 | thedotmack/claude-mem | +4,473 | 115,867 | 2025-08-31 | TypeScript | Apache-2.0 |
| 10 | Donchitos/Claude-Code-Game-Studios | +3,730 | 38,099 | 2026-02-12 | Shell | MIT |
Top New Repos — Top 10
| # | Repository | Total stars | Created | Language | License |
|---|---|---|---|---|---|
| 1 | NotRexed/Hookshot | 6,652 | 2026-11-05 | C++ | GPL-3.0 |
| 2 | huamidev/GestaltX | 5,007 | 2026-11-09 | Swift | GPL-3.0 |
| 3 | tangerineturtle1/NX2navi | 2,917 | 2026-11-07 | C | MIT |
| 4 | jhaizhou-ops/S2Arm | 2,412 | 2026-11-08 | C# | Apache-2.0 |
| 5 | JenkinWoo/sheeridVerify | 2,010 | 2026-11-06 | JavaScript | ISC |
| 6 | IainKay/ram-altman | 1,829 | 2026-11-05 | Shell | Unlicense |
| 7 | jamesondh/chimera-vocal-synth | 1,718 | 2026-11-05 | Rust | MIT |
| 8 | DIYgod/pinky | 1,709 | 2026-11-06 | Go | Apache-2.0 |
| 9 | mmdoctor/google-pro-helper | 1,654 | 2026-11-05 | Python | Undeclared |
| 10 | livliv-dev/player | 1,465 | 2026-11-04 | TypeScript | Apache-2.0 |
Projects worth a closer look
T3 Code: the workspace becomes the product
Its official docs combine threads, permission modes, source control, remote access, and background work. Its 11,979 new stars suggest growing interest in long-running work, not just one-shot answers. An HN discussion shows interest alongside questions about synchronization and maturity, so test it on a disposable repository before handing it sensitive or irreversible work.
security-audit-skill: audit results as re-checkable artifacts
Cloudflare describes a multi-phase audit with independently verified, machine-readable findings. That is attractive because a finding can leave evidence instead of becoming another prompt transcript. It still cannot decide whether a finding crosses your own trust boundary. Validate it with harmless fixtures before connecting it to a real pipeline.
agent-browser and agent-desktop: from web pages to desktops
The agent-browser repository describes a native Rust CLI that drives a browser through an accessibility tree; agent-desktop extends computer use toward desktop control. This is easy to imagine for travel planning and data chores, but logged-in accounts, payments, and deletion actions should stay out of an unverified agent. Start with allowlisted domains, test accounts, and a recovery path.
Scrapling: data access needs product-level boundaries
Scrapling combines HTTP, browser, spider, MCP, and Agent Skill paths in one Python toolkit. That is useful for research, but targets, proxies, site terms, and personal data in exports need separate controls. Community interest cannot replace authorization and scope checks.
claude-mem: memory is not free convenience
It carries compressed session observations into future work, which is compelling for long projects. It also means prompts, tool output, and provider data flows need an understandable lifecycle. An HN discussion raised concrete privacy concerns. Try it in a test profile first, confirm consent and deletion, then decide whether it belongs in daily work.
What the community says
The most useful opposing signals this week come from two places. The T3 Code HN discussion shows user interest in workspace and agent collaboration, but does not prove reliability. A claude-mem HN discussion raises concrete questions about prompt history and upload consent. The first shows demand forming; the second reminds us to evaluate convenience with data lifecycle.
Editor's picks
Try now: Test T3 Code, security-audit-skill, or agent-browser in an isolated repository with least privilege and a saved recovery path.
Watch first: claude-mem, agent-desktop, Scrapling, and the new hardware or platform projects. Their direction is interesting, but data flow, compatibility, and failure handling need evidence.
Do not adopt directly: Anything that bypasses platform rules, handles third-party verification, uses real login sessions, or requests unclear permissions. Hookshot and GestaltX being popular does not establish maturity or authorization for your use case.
Was this article helpful?



