GitHub Weekly Vol.5: 6 Projects Worth a Look This Week

★ 190,486·Python·read 11-10

Anthropic's official Agent Skills example repo: 19 skill folders spanning creative work, development, enterprise comms and document handling — the docx, pdf, pptx and xlsx skills are the same implementations behind Claude's document features. Some skills are Apache 2.0.

Why look

4th on weekly trending; the official line is simple: a skill is a folder plus a SKILL.md, installable in Claude Code with one marketplace command; the repo also ships the Agent Skills spec and a template.

The catch

Not tested this issue: the README states these are for demonstration and learning only, and to test before relying on them for critical tasks; there is no LICENSE file at the repo root, so check each skill's licence before commercial use.

Best for

Anyone writing skills for Claude, or studying how the official document skills are built; read the disclaimer before copying them into production.

Open on GitHub →

★ 148,941·Python·read 11-10

GitHub's Spec-Driven Development toolkit: write the spec first, then let a coding agent walk constitution, spec, plan, tasks, implementation and convergence, with bug-fixing and idea-assessment flows included. MIT.

Why look

14th on weekly trending; tested this issue: specify 1.3.2 scaffolded a project in 1.3 seconds with templates, scripts, workflows and Copilot integration in one pass.

The catch

The full six-step flow runs inside an agent chat, one command at a time — this run verified scaffolding only, not a complete feature build; the README asks for Python 3.11+ and uv, while this run installed the same CLI via pip.

Best for

Anyone who wants agents to build from a spec instead of a one-line prompt; existing projects can adopt it without starting over.

Open on GitHub →

★ 100,369·Python·read 11-10

A Python scraping framework: fast requests, stealth browsers, JS rendering, concurrent spiders, proxy rotation and pause/resume in one library, plus a CLI, MCP server and agent skill. BSD-3-Clause.

Why look

10th on weekly trending; this issue ran the README's own example: one request returned 10 quotes and 10 authors, status 200, in 1.7 seconds.

The catch

The base install alone does not run: Fetcher needs the fetchers extra (curl_cffi, patchright, browserforge and more), installed separately in this run; browser rendering needs browser binaries, not tested here.

Best for

Anyone scraping static pages without hand-rolling headers and parsers per site; for JS rendering or anti-bot cases, check the fetcher comparison in the docs first.

Open on GitHub →

★ 37,399·Shell·read 11-10

Turns one Claude Code session into a full game dev studio: 49 agents in three tiers (directors, department leads, specialists) with 74 skills, 15 hooks and 14 rules, covering Godot 4, Unity and Unreal 5. MIT.

Why look

5th on weekly trending; this issue cloned and counted: 49 agents, 74 skills and 14 rules all match the README, with 18 scripts in the hooks directory; the three directors default to Opus while most of the studio inherits the session model.

The catch

Not tested this issue: the full flow needs a real game project inside Claude Code, and the docs themselves ask first whether the output is a game or documents; the coordination cost of 49 agents is unverified.

Best for

Anyone building games with Claude Code who misses studio structure; for a quick prototype, start with /quick-design.

Open on GitHub →

★ 53,609·TypeScript·read 11-10

A remote control surface for agent harnesses: phone, web and desktop apps drive Claude Code, Codex, Cursor, OpenCode, Antigravity, Grok Build and fx on your machine, on your own subscriptions. MIT.

Why look

7th on weekly trending; this issue cloned it and confirmed a large monorepo — apps/server, web and desktop plus dozens of provider packages; one command installs it, and npx t3@latest tries it without installing.

The catch

Not tested this issue: full use needs at least one provider CLI installed and logged in on the machine; the README calls itself very early, expects bugs, and is mostly not accepting contributions yet.

Best for

Anyone who checks on agents from their phone; for a one-off look without a background service, try the npx run first.

Open on GitHub →

★ 17,248·TypeScript·read 11-10

An open-source Zi Wei Dou Shu charting engine built on Ni Haixia's Tian Ji system: full charting algorithms, the four-transformations system, pattern knowledge base and classical texts, plus a v4.0 dataset of 674,976 charts × 13 dimensions. Code MIT, dataset CC BY 4.0.

Why look

9th on weekly trending; tested this issue: npm install took 16 seconds for 184 packages, npm test passed in 1 minute 14 seconds, and typecheck finished clean in 8 seconds; the 56.6 GB unpacked dataset size is the author's figure, not downloaded here.

The catch

The engine ships without a backend: AI interpretation endpoints must be implemented yourself, and the open build says so plainly when called; the v4.0 dataset was generated under the old default brightness school, with small differences from the new default, as the README notes.

Best for

Anyone studying Zi Wei Dou Shu algorithms or embedding charting in their own product; for the interpretation text alone, download the dataset instead of running the engine.

Open on GitHub →