GitHub Weekly Vol.6: 7 Projects Worth a Look This Week
7 open-source projects worth your time this week: 3 tested hands-on, 4 code-reviewed. Meta’s design system tops the chart, and a username-lookup tool from 2018 is flowing back.

Latest
1 updates · newest first-
Weekly debut: GitHub Weekly Vol.6, 7 projects worth a look this week
Further updates will appear here
Vol.6: A design system tops the chart, an old tool flows back
This week the top of GitHub trending belonged to agent tooling and design systems: Meta’s Astryx gained about 13,501 stars to take the top spot, while sherlock from 2018 trended for a second straight week; in the same week, Google’s design-review plugin, Vercel’s agent browser, Addy Osmani’s engineering skills and a Chinese stock-analysis system all climbed. 7 projects this issue, 3 tested hands-on, 4 code-reviewed — only verified honest takes.
At a glance
★ tracked by us daily| # | Project | In one line | How | ★ | Week |
|---|---|---|---|---|---|
| 01 | meta/astryx | Meta’s in-house design system of 8 years, open-sourced: 150+ accessible components, 7 customizable themes and a CLI, built on React 19 and StyleX, MIT. | Tested | 29,433 | — |
| 02 | sherlock-project/sherlock | A command-line tool from 2018: give it a username and it checks 400+ social networks and sites for where it is registered. MIT. | Tested | 112,990 | — |
| 03 | google/elephant-goldfish-model | Google’s open-source plugin: a design-review workflow for coding agents — the Elephant holds project context and drafts the spec, Goldfish reviewers read only the document to find blind spots. 7 skills, Apache 2.0. | Read | 20,207 | — |
| 04 | ZhuLinsen/daily_stock_analysis | A Chinese open-source stock analysis system covering A-shares, Hong Kong, US, Japanese, Korean and Taiwan stocks plus ETFs: daily decision dashboards pushed to WeChat Work, Feishu, Telegram and more. MIT. | Read | 72,741 | — |
| 05 | vercel-labs/agent-browser | Vercel Labs’ browser-automation CLI: a browser for AI agents — open a page with one command, snapshot reads it as an accessibility tree with refs like @e1, then click and fill by ref. Native Rust, Apache 2.0. | Tested | 51,907 | — |
| 06 | addyosmani/agent-skills | Addy Osmani’s production-grade engineering skills: 25 skills covering the full dev lifecycle from spec to ship, with 9 slash commands. MIT. | Read | 118,210 | — |
| 07 | hugohe3/ppt-master | AI turns documents or topics into native PowerPoint decks that stay editable — slide masters, native shapes, data-backed charts and tables, not filled templates or flat images. Runs as a skill inside any agent-capable tool. MIT. | Read | 77,146 | +1,911 |
One by one
All issues-
Meta’s in-house design system of 8 years, open-sourced: 150+ accessible components, 7 customizable themes and a CLI, built on React 19 and StyleX, MIT.
- Why look
- No. 1 on weekly trending, up about 13,501 stars in a week; tested this issue: CLI 0.7.16 installed globally in 28 seconds,
component --listprinted 174 components, matching the README’s 150+; three patch releases shipped Nov 11–14. - The catch
- This run verified the packages load and the CLI runs, but did not render a React component in a browser; still in beta, so the API may change.
- Best for
- Anyone building product UI on React 19, or studying how Meta organizes its in-house design system; run your own rendering check before shipping it to production.
How we tested
On Node 24.22.1, npm installed @astryxdesign/cli 0.7.16 globally in 28 seconds (142 packages);
astryx component --listprinted 174 components in about 20 seconds; a minimal project installed @astryxdesign/core, @astryxdesign/theme-neutral and @stylexjs/stylex in 10 seconds, core loaded with 532 exports in Node and the Button subpath import worked. -
A command-line tool from 2018: give it a username and it checks 400+ social networks and sites for where it is registered. MIT.
- Why look
- No. 3 on weekly trending, up about 10,412 stars in a week and trending for a second straight week, with no new release in over two months — a word-of-mouth comeback; tested this issue: pip installed 0.16.2 in 1 minute 20 seconds, a lookup for octocat hit 83 sites, and the bundled site list measured 481 sites, so the 400+ claim holds.
- The catch
- Full runs can miss real accounts: this run’s full sweep missed the real GitHub account, found only on a site-limited re-run; Reddit is 403-blocked from this machine’s egress IP, so a nonexistent username was also reported as found there — re-check key sites individually.
- Best for
- Anyone checking where their own username is registered, or starting out with OSINT; re-check key sites individually before treating a hit as evidence.
How we tested
On Python 3.12.3, pip installed sherlock-project 0.16.2 in a venv in 1 minute 20 seconds;
sherlock octocatswept 481 sites in about 2 minutes with 83 hits; a --site GitHub, GitLab, Reddit re-check took 6.7 seconds (GitHub and Reddit hit, GitLab missed); a fake username, himiao-test-xyz-12345, across 5 sites took 4.9 seconds with 1 false positive on Reddit. -
Google’s open-source plugin: a design-review workflow for coding agents — the Elephant holds project context and drafts the spec, Goldfish reviewers read only the document to find blind spots. 7 skills, Apache 2.0.
- Why look
- No. 9 on weekly trending, up about 5,060 stars in a week; all 7 skills checked one by one, each with its SKILL.md, covering brainstorming, drafting, review, review loops and drift reports; reviews need an agent that can open fresh subagent sessions, an idea from Google engineer Dave Rensin’s essay.
- The catch
- Not tested this issue: the full flow runs inside an agent session in Claude Code, Codex or Antigravity, so this run verified only the skill structure and docs; the repo states it is not an officially supported Google product.
- Best for
- Anyone whose design docs keep getting blind spots flagged in review; start with egm-review on an existing document instead of running the full flow.
Read
Read only: not installed or run — we read the code and docs
-
A Chinese open-source stock analysis system covering A-shares, Hong Kong, US, Japanese, Korean and Taiwan stocks plus ETFs: daily decision dashboards pushed to WeChat Work, Feishu, Telegram and more. MIT.
- Why look
- No. 6 on weekly trending, up about 4,548 stars in a week to about 72,500 stars, with about 57,600 forks — nearly as many forks as stars, and since the workflow is fork-then-configure-your-own-keys, many of those forks are self-hosted deployments; the repo ships 309 test files and links an arXiv paper.
- The catch
- Not tested this issue: full analysis needs a model API key and market-data sources; the free sources claim zero-config operation but with no stability guarantee; the README top carries sponsor slots from several API resellers — keep that in mind when picking a model service.
- Best for
- Anyone wanting to self-host a daily stock-analysis feed; read the data-source config docs first — the free market-data sources are for trying out.
Read
Read only: not installed or run — we read the code and docs
-
Vercel Labs’ browser-automation CLI: a browser for AI agents — open a page with one command, snapshot reads it as an accessibility tree with refs like @e1, then click and fill by ref. Native Rust, Apache 2.0.
- Why look
- No. 5 on weekly trending, up about 4,217 stars in a week; tested this issue: agent-browser 0.43.0 installed globally in 24 seconds, Chrome for Testing 157.0.8103.78 (194 MB) downloaded, example.com and GitHub trending really opened, snapshots in milliseconds, clicking @e1 navigated correctly; ships 10 skills plus an MCP server.
- The catch
- Fails out of the box on networks with an authenticated proxy plus TLS interception: the documented proxy auth did not work in this run — a local forwarding proxy plus --ca-cert was needed, and every command must repeat both flags or the session resets; the Chrome download stalled twice at 80% before succeeding on the third try.
- Best for
- Anyone giving an agent a browser it can click through; works out of the box on a direct connection — on a restricted network, set up the proxy and certificates first.
How we tested
On Node 24.22.1, npm installed agent-browser 0.43.0 globally in 24 seconds;
agent-browser installdownloaded Chrome for Testing 157.0.8103.78 (194 MB) in 4 minutes 30 seconds with two stalled retries; via a local forwarding proxy plus --ca-cert, example.com opened in 2.1 seconds, snapshot found the Learn more link @e1, clicking it landed on iana.org, GitHub trending opened in 10.4 seconds with a full accessibility-tree snapshot, and close exited cleanly. -
Addy Osmani’s production-grade engineering skills: 25 skills covering the full dev lifecycle from spec to ship, with 9 slash commands. MIT.
- Why look
- No. 11 on weekly trending, up about 3,889 stars in a week to about 117,900 stars; this run cloned and counted the skills directory — exactly 25, matching the README; one npx command installs into 70+ agents, and the Claude Code plugin marketplace is supported.
- The catch
- Not tested this issue: skills are docs and workflows for agents to read, not programs that run directly; the README notes that installing a single skill leaves its referenced shared checklists behind, tracked in issue #361.
- Best for
- Anyone who wants agents to build from a fixed workflow instead of a one-line prompt; install the whole repo first, then pick individual skills as needed.
Read
Read only: not installed or run — we read the code and docs
-
AI turns documents or topics into native PowerPoint decks that stay editable — slide masters, native shapes, data-backed charts and tables, not filled templates or flat images. Runs as a skill inside any agent-capable tool. MIT.
- Why look
- Up about 2,282 stars this week on the site watchlist, to about 77,200 stars, with only 1 open issue; this run cloned and checked it: skills/ppt-master ships SKILL.md, scripts, templates, workflows and references, structure complete; the author is blunt in the README — this is a tool, not a wishing well, and the model sets the ceiling.
- The catch
- No full deck was generated this issue: generation runs inside an agent tool and needs your own model API key (Kimi K3.1 or a large-context Gemini is recommended), so this run verified only the repo structure and docs; the author notes that cheap models will not produce good decks — upgrade the model before debugging usage.
- Best for
- Anyone who regularly turns PDFs or source material into a PPT they can keep editing; have a model key and a large-context model ready first.
Read
Read only: not installed or run — we read the code and docs
Why it matters
these 7 projects span design systems, browser automation, username lookup, skills, design review, slide generation and stock analysis, as agent tooling keeps moving from single prompts to full workflows.
Follow HIMIAO Horizon
No sign-up · Unsubscribe any timeNotifications
A ping on your phone or computer when the brief is out. No email needed.
Comments
Related
- Nov 9GitHub Weekly Vol.5: 6 Projects Worth a Look This Week
- Nov 2GitHub Weekly Vol.4: 6 Projects Worth a Look This Week
- Nov 2DeepSeek's Super-Agent, Explained: A Router Dispatches, Workers Execute, and It Speaks Only When Allowed
- Oct 26GitHub Weekly Vol.3: 6 Projects Worth a Look This Week
- Oct 20Google Open-Sources Harness²: Letting AI Agents Rewrite Their Own Scaffolding, Hits GitHub Trending
Today
Nov 17 Tuesday- Brief
- GitHubGitHub Weekly Vol.6: 7 Projects Worth a Look This Week
Tested or read — weeklyAll →
- Quiz
- Call
Latest news
All →- Nov 17OpenAI Launches Science Blog as GPT-6 Astra Leads Proof of Candidate Lattice Standard Model Definition
- Nov 17Apple Smart Glasses Said to Debut at WWDC 2027 With No Display, Leaning on Siri and iPhone
- Nov 17AI Labs Race to Hire Counterintelligence Staff: Anthropic 33 This Year, OpenAI 22
- Nov 17Trump Backs Huang's Return to Nvidia: 'He Learned a Big Lesson'
- Nov 16Kuo: OpenAI and Anthropic to Buy About 26M Google TPUs in 2028–29
- Nov 16Huawei Cloud Launches Ascend-Native Embodied World Model CloudRobo-WM1.2
™