杜南风

杜南风

跑步数据攒了不少,想看看有没有什么有意思的变化。先做给自己用。

Joined 10/30/2026
2
Projects
0
Collaboration requests
0
Iterated
0
Likes
0
Saves
6
Views
0
Comments
Toolsclaude-code

Yab

A small, quiet Mac browser with Arc's spirit and agents in their own tabs: WebKit or Chromium per tab, and your private tabs stay private.

Yab is a small, quiet browser for the Mac, with an iPhone app in beta. Its pitch is plain: keep what people loved about Arc, and add hands — your own Claude or ChatGPT doing the work in your tabs. It is free, at version 0.50.0, for macOS 15.4 or later. Every tab or site runs on WebKit or, on Apple silicon, on Chromium, with Chrome extensions, DevTools and the Chrome Web Store. The released app opens pages in Chromium by default and uses WebKit for protected video and Apple Pay checkouts. An ad blocker runs inside the engine before a request is made, and reading mode, floating video, split view and Peek are each one key away. The agents are the point. Agents from the AI providers you already pay for work in tabs of their own, with your sign-ins, asking before they send, pay or delete — and they never see a private tab. Sync between your Macs and iPhone is optional and end-to-end encrypted, and passwords live in your keychain or the manager you already use. It has 738 stars on GitHub with 75 forks, was created on 14 November 2026 and was last pushed on 20 November 2026. It is Apache-2.0 licensed. If you would rather build it yourself, Xcode 26 or later plus a build script makes a separate Yab Dev copy that runs beside the real app without touching it.

Security & Privacyclaude-code

Penthera

An offline security agent for your app, your dependencies, and your AI coding tools — it scans, grades, and fixes on your own machine.

Penthera is a free, offline security checker for your own app, its dependencies, and the AI coding tools you build with. It runs on your own machine — no install, no account — and gives you a grade per area plus the three things to fix first. It checks a lot: your running app or preview URL, your repository, your dependencies, and your AI tooling. Every finding comes from a rule you can read, with its evidence at file:line — a model may propose a problem, but a deterministic rule decides whether it counts. It does more than point: it applies fix recipes on a local branch, re-scans to prove they worked, and never pushes anything. There is a local dashboard with a Simple mode that states the verdict in one plain sentence, and it plugs into your coding agent through an MCP server, a Claude Code plugin, Codex, Copilot, or Gemini CLI. It has 104 stars on GitHub and 1,610 commits behind it, at version 2.4.0. Two honest limits: it finds known classes of problems, not new ones, and it does not replace a human pentester. Its AI mode is still experimental and unmeasured, and the live check on penthera.org is beta and passive only. It is open R&D — every check is measured in public, and the results are published, good or bad.