GitHub Open Source Weekly 2026-05-27: Code Knowledge Graphs Dominate, Skills Ecosystem Goes Official, Supply Chain Security Strikes Back
Data window: 2026-05-19 to 2026-05-27 (rolling 7 days) Sources: GitHub Trending weekly + monthly, GitHub Search API, HN Algolia
TL;DR: This week's defining theme is "how AI agents read and understand codebases." codegraph (+20,208 stars) and Understand-Anything (+14,750 stars, 169 HN upvotes, 49 comments) both topped the charts simultaneously — signaling that developers are starting to treat code knowledge graphs as essential infrastructure. The Skills ecosystem expanded from individual repos (academic-research-skills) to enterprise platforms (dotnet/skills), with Anthropic's knowledge-work-plugins making its first appearance on the list. The surprise entry: Perplexity AI officially released Bumblebee, a supply chain security scanner for developers — a sign that AI companies are moving beyond models into security tooling.
📈 Fastest Growing — Weekly Star Gain Top 15
Source:
github.com/trending?since=weekly🔁 = also appearing on monthly trending (sustained interest signal)
| # | Repo | +Stars/week | Total Stars | Language | Created |
|---|---|---|---|---|---|
| #1 | 🔁 colbymchenry/codegraph | +20,208 | 27,607 | TypeScript | 2026-01-18 |
| #2 | 🔁 Lum1104/Understand-Anything | +14,750 | 35,615 | TypeScript | 2026-03-15 |
| #3 | tinyhumansai/openhuman | +11,906 | 28,294 | Rust | 2026-02-18 |
| #4 | 🔁 Imbad0202/academic-research-skills | +10,678 | 22,134 | Python | 2026-02-26 |
| #5 | 🔁 rohitg00/ai-engineering-from-scratch | +10,035 | 20,635 | Python | 2026-03-18 |
| #6 | ruvnet/RuView | +6,396 | 66,303 | Rust | 2025-06-07 |
| #7 | 🔁 rohitg00/agentmemory | +5,687 | 18,202 | TypeScript | 2026-02-25 |
| #8 | HKUDS/CLI-Anything | +4,010 | 40,610 | Python | 2026-03-08 |
| #9 | HKUDS/ViMax | +2,790 | 7,623 | Python | 2025-03-30 |
| #10 | anthropics/knowledge-work-plugins | +2,666 | 16,620 | Python | 2026-01-23 |
| #11 | can1357/oh-my-pi | +2,584 | 7,521 | TypeScript | 2025-12-31 |
| #12 | supertone-inc/supertonic | +2,329 | 10,633 | Swift | 2025-11-18 |
| #13 | humanlayer/12-factor-agents | +1,985 | 22,413 | TypeScript | 2025-03-30 |
| #14 | presenton/presenton | +1,787 | 7,068 | TypeScript | 2025-05-10 |
| #15 | dotnet/skills | +1,313 | 3,108 | C# | 2026-02-03 |
🆕 Top New Repos — This Week's Newcomers Top 10
Source: GitHub Search API (
created:2026-05-19..2026-05-27, sorted by total stars)
| # | Repo | Total Stars | Language | Created |
|---|---|---|---|---|
| #1 | perplexityai/bumblebee | 3,156 | Go | 2026-05-20 |
| #2 | FoundZiGu/GuJumpgate | 2,691 | JavaScript | 2026-05-19 |
| #3 | thananon/9arm-skills | 2,342 | Shell | 2026-05-20 |
| #4 | open-gsd/get-shit-done-redux | 1,083 | JavaScript | 2026-05-22 |
| #5 | Tong89/smartNode | 1,077 | Python | 2026-05-21 |
| #6 | run-liyi/wechatpay | 770 | JavaScript | 2026-05-21 |
| #7 | MoonshotAI/kimi-code | 713 | TypeScript | 2026-05-22 |
| #8 | kageroumado/phosphene | 686 | Swift | 2026-05-20 |
| #9 | 0xSero/codex-shim | 635 | Python | 2026-05-22 |
| #10 | VILA-Lab/FigMirror | 309 | Python | 2026-05-22 |
Weekly Spotlight — Fastest Growing Top 15
📈 #1 — colbymchenry/codegraph｜Pre-indexed code knowledge graph for AI agents, runs fully local
Pre-indexed code knowledge graph for Claude Code, Codex, Gemini, Cursor, OpenCode, AntiGravity, and Hermes Agent — fewer tokens, fewer tool calls, 100% local
+20,208 ★ this week｜27,607 total｜TypeScript｜MIT｜Monthly trending
codegraph tackles a concrete problem: when you drop a large repo into an AI agent, the agent typically burns through dozens of tool calls just to map the codebase — slow and token-expensive. codegraph pre-parses the entire codebase into a semantic knowledge graph, giving Claude Code, Codex, Cursor, and other agents structured context before they ever make a tool call. Fewer tokens, fewer round trips.
The repo's explosive growth comes down to a few factors. First, it claims support for nearly every major AI coding agent (7 listed in the description) rather than being tied to one tool. Second, 100% local execution is a meaningful differentiator for enterprise users who can't send code to third-party services. The 170 open issues signal rapid community adoption — and the maintenance pressure that comes with it.
📈 #2 — Lum1104/Understand-Anything｜Interactive code knowledge graph with HN community validation
Graphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
+14,750 ★ this week｜35,615 total｜TypeScript｜MIT｜Monthly trending
Also a code knowledge graph, but with a distinct positioning: Understand-Anything emphasizes interactivity — you can search, explore, and query the graph directly, not just use it as background context for an agent. The 169-point HN thread with 49 comments is the highest-quality discussion in this week's roundup, with the core debate being whether "graph-first" approaches genuinely outperform traditional RAG.
The fact that codegraph and Understand-Anything both surged the same week isn't coincidence — they're solving the same engineering problem from slightly different angles. That convergence is the signal.
📈 #3 — tinyhumansai/openhuman｜Personal AI super-intelligence in Rust, GPL-3.0
Your Personal AI super intelligence. Private, Simple and extremely powerful.
+11,906 ★ this week｜28,294 total｜Rust｜GPL-3.0
openhuman positions itself as a personal AI super-intelligence: private, local, Rust-based. The tinyhumans.ai/openhuman website emphasizes three properties — Private, Simple, Powerful. Missing topics and a vague description suggest this is still early-stage, but the Rust + privacy-first combination has a clear community appetite.
The GPL-3.0 license is worth noting. Unlike MIT or Apache, GPL-3.0's copyleft "infection" restricts commercial closed-source use. Choosing GPL-3.0 is often a deliberate signal that the author wants to prevent proprietary forks.
📈 #4 — Imbad0202/academic-research-skills｜Full academic research workflow for Claude Code, 82 HN points
Academic Research Skills for Claude Code: research → write → review → revise → finalize
+10,678 ★ this week｜22,134 total｜Python｜Monthly trending
A complete Claude Code skills suite for academic research, covering the full pipeline from literature review through writing, peer review, and final revision. The 82-point HN thread with 25 comments is the second-highest quality discussion this week — centered on what role AI agents should play in academic writing. Supporters frame it as a productivity tool; critics raise academic integrity concerns.
Together with dotnet/skills (#15), this repo signals that skills verticalization is accelerating: rather than generic agent capabilities, the community is building deep specializations for specific professional contexts.
📈 #5 — rohitg00/ai-engineering-from-scratch｜Comprehensive AI engineering curriculum, weekly updates
Learn it. Build it. Ship it for others.
+10,035 ★ this week｜20,635 total｜Python｜MIT｜Monthly trending
A structured AI engineering learning repo covering agents, MCP, RAG, transformers, RL, and most of what an AI engineer needs today. Primary language is Python, with TypeScript and Rust content as well. The official site is aiengineeringfromscratch.com. Resources like this with consistent weekly updates and a clear learning arc tend to accumulate stars slowly but reliably — the monthly trending co-appearance confirms a stable follower base.
If you're mapping out an AI engineering learning path, or need a go-to recommendation for a team member getting started, the topics (agents, ai-engineering, swarm-intelligence, mcp) give you a clear sense of what's covered.
📈 #6 — ruvnet/RuView｜WiFi signals → real-time spatial awareness, no cameras required
π RuView turns commodity WiFi signals into real-time spatial intelligence, vital sign monitoring, and presence detection — all without a single pixel of video.
+6,396 ★ this week｜66,303 total｜Rust｜MIT
RuView is this week's most counterintuitive entry: no cameras, just existing WiFi signals — yet it can detect human position, vital signs, and posture. The technology is DensePose algorithm applied to RF signal analysis, with support for ESP32 MCUs and Home Assistant integration.
With 66,303 total stars and 8,787 forks, this repo already has substantial reach. The +6,396 this week represents a new wave of discovery. The privacy angle is real in both directions: for users who want camera-free monitoring, WiFi sensing is appealing — but any device that can read your WiFi signals can do the same thing.
📈 #7 — rohitg00/agentmemory｜Persistent memory for AI coding agents across sessions
#1 Persistent memory for AI coding agents based on real-world benchmarks
+5,687 ★ this week｜18,202 total｜TypeScript｜Apache-2.0｜Monthly trending
agentmemory solves the amnesia problem: every new session, your AI coding agent forgets everything it learned last time. This repo provides a persistent memory layer across sessions, supporting Claude, Codex, Cursor, Copilot, and other major tools. Monthly chart co-appearance suggests it's built consistent community trust.
The topics reveal broader ecosystem ambitions: integration with agentmemory, harness, and hermes signals this is positioned as foundational agent infrastructure, not a standalone utility.
📈 #8 — HKUDS/CLI-Anything｜Make any CLI tool agent-native
"CLI-Anything: Making ALL Software Agent-Native"
+4,010 ★ this week｜40,610 total｜Python｜Apache-2.0
CLI-Anything has an ambitious goal: let AI agents operate any CLI tool directly, without needing to write individual MCP servers or API wrappers for each one. Official site at clianything.cc. With 40,610 total stars, this repo has been accumulating traction for months — the +4,010 this week is continued steady growth rather than a spike.
If your workflow relies on many CLI tools (git, aws, kubectl, etc.), this approach is worth tracking. Universal agent-to-CLI bridging is more scalable than per-tool integration.
📈 #9 — HKUDS/ViMax｜Autonomous AI video generation — director, screenwriter, producer in one
"ViMax: Agentic Video Generation (Director, Screenwriter, Producer, and Video Generator All-in-One)"
+2,790 ★ this week｜7,623 total｜Python｜MIT
ViMax, also from HKUDS, decomposes video generation into four agent roles: director (scene planning), screenwriter (script generation), producer (resource orchestration), and renderer (actual video generation). Together they form a fully autonomous video production pipeline. Alongside AIDC-AI/Pixelle-Video on the monthly chart, ViMax represents where agentic AI meets multimedia creation.
📈 #10 — anthropics/knowledge-work-plugins｜Anthropic's official plugins for Claude Cowork knowledge workers
Open source repository of plugins primarily intended for knowledge workers to use in Claude Cowork
+2,666 ★ this week｜16,620 total｜Python｜Apache-2.0
This is Anthropic's official plugin repo, designed specifically for knowledge workers in Claude Cowork environments. Its presence alongside anthropics/financial-services (monthly chart #3) shows Anthropic is building out vertical-specific official plugin strategies across multiple domains.
For developers and enterprise users building on Claude: official Anthropic repos often preview platform feature direction before those features land in the API. Worth watching.
📈 #11 — can1357/oh-my-pi｜Terminal AI coding agent with hash-anchored edits, LSP, and subagents
AI Coding agent for the terminal — hash-anchored edits, optimized tool harness, LSP, Python, browser, subagents, and more
+2,584 ★ this week｜7,521 total｜TypeScript｜MIT
oh-my-pi (site: omp.sh) is a terminal AI coding agent that differentiates itself through hash-anchored edits (precise file changes that don't drift with line numbers), LSP integration, multi-provider support (Claude + OpenAI), and a subagent system. Built with Bun + TypeScript. The 204 open issues signal active community use — and a healthy backlog of bugs to chase.
In a crowded terminal AI agent space, oh-my-pi finds its niche through "precise editing" and "subagent collaboration."
📈 #12 — supertone-inc/supertonic｜On-device multilingual TTS via ONNX, native Swift
Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.
+2,329 ★ this week｜10,633 total｜Swift｜MIT
supertonic is this week's most unexpected entry — and most distinctly non-agent-framework repo. Supertone is a well-known Korean voice AI company (known for AI vocal recreation for K-pop artists). They've open-sourced their multilingual TTS engine as Swift + ONNX, running natively on-device (including iOS) without cloud API dependency. Official demo on HuggingFace Spaces (supertonic-3).
Supported languages include Chinese, English, and Japanese. Supported platforms span iOS, Python, Node.js, Go, Flutter, C++, and WebGPU. For developers who need embedded TTS without third-party API dependencies, this is a serious option.
📈 #13 — humanlayer/12-factor-agents｜12 design principles for production-ready LLM agents
What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers?
+1,985 ★ this week｜22,413 total｜TypeScript
Inspired by the 12-factor app methodology, 12-factor-agents systematizes the design principles for production-ready LLM agents — covering context window management, memory, RAG, orchestration, and prompt engineering. Created in March 2025 and still gaining stars consistently, which means it's filling the gap in documentation on "how to move an agent from demo to production."
If your agent system is making the leap from prototype to prod, this is a useful design framework to work through systematically.
📈 #14 — presenton/presenton｜Open-source AI presentation generator, Gamma alternative
Open-Source AI Presentation Generator and API (Gamma, Beautiful AI, Decktopus Alternative)
+1,787 ★ this week｜7,068 total｜TypeScript｜Apache-2.0
presenton positions itself as an open-source alternative to Gamma — AI-generated PowerPoint/slides with a programmatic API. Official site at presenton.ai. The steady +1,787 growth shows sustained demand for AI presentation generation in the vertical, though open-source solutions still lag behind Gamma and Beautiful AI on ease of use and visual polish.
📈 #15 — dotnet/skills｜Microsoft's official .NET and C# skills for AI coding agents
Repository for skills to assist AI coding agents with .NET and C#
+1,313 ★ this week｜3,108 total｜C#｜MIT
This is a skills repo under Microsoft's official .NET org, providing language-specific skills for AI coding agents working with .NET and C#. The significance here isn't technical novelty — it's the signal: when the official dotnet org starts maintaining an agent-skills repo, that paradigm has entered mainstream engineering culture.
Alongside this week's academic-research-skills and anthropics/knowledge-work-plugins, skills are clearly expanding from individual developers to large organizations and official maintainers.
Weekly Spotlight — Top New Repos
🆕 #1 — perplexityai/bumblebee｜Perplexity's official supply chain security scanner for developers
Read-only developer endpoint scanner for on-disk package, extension, and developer-tool metadata, built to check exposure to known software supply-chain compromises.
3,156 total ★｜Go｜Apache-2.0｜Created 2026-05-20
bumblebee is Perplexity AI's official open-source developer security tool. It performs read-only scans of packages, extensions, and developer tool metadata on a developer's machine, cross-referencing against known software supply chain compromises. Written in Go, supports macOS and Linux.
The technical complexity here is modest, but the strategic timing is interesting. AI company toolchains — MCP servers, agent plugins, coding assistants — are becoming a new supply chain attack surface. Perplexity building this tool internally and open-sourcing it likely reflects real internal security needs. Developers who are actively using AI coding tools should be more systematic about tracking dependency exposure in their environments.
🆕 #2 — FoundZiGu/GuJumpgate｜2,691 stars, no description
2,691 total ★｜JavaScript｜MIT｜Created 2026-05-19
GuJumpgate has no repo description, but accumulated 2,691 stars and 727 forks in 5 days (the fork-to-star ratio is unusually high — worth noting). Without verifiable content, treat with caution.
🆕 #3 — thananon/9arm-skills｜Shell-based skills, 2,342 stars
2,342 total ★｜Shell｜Created 2026-05-20
No description, 323 forks, written in shell scripts. "9arm" may reference Thai developer thananon's brand, but without more information the content quality can't be assessed.
🆕 #4 — open-gsd/get-shit-done-redux｜Claude Code context engineering framework
Getting Shit Done, the Aftermath
1,083 total ★｜JavaScript｜MIT｜Created 2026-05-22
A community fork of the original get-shit-done framework, focused on Claude Code context engineering and spec-driven development. Topics include meta-prompting and claude-code, making this a practical workflow framework rather than theoretical documentation. 1,083 stars in 5 days signals clear community demand for this approach.
🆕 #5 — MoonshotAI/kimi-code｜Moonshot AI's official agent framework
The Starting Point for Next-Gen Agents
713 total ★｜TypeScript｜MIT｜Created 2026-05-22
This is the official agent framework from Moonshot AI (Kimi), positioned as "the starting point for next-generation agents." Official site at moonshotai.github.io/kimi-code. At 713 stars in 5 days from a top-tier Chinese AI lab, significant growth ahead is likely. Its appearance on the new repo chart is this week's clearest "major AI companies are open-sourcing their agent infrastructure" signal.
🆕 #6 — Other new repos worth noting
- kageroumado/phosphene (686 stars, Swift, macOS Tahoe dynamic wallpaper engine): Developer tooling for the upcoming macOS version is already appearing — the community moves fast.
- 0xSero/codex-shim (635 stars, Python): Lets Codex Desktop connect to Factory BYOK models and GPT-5.5. A classic "wrap one API for another tool" utility.
- VILA-Lab/FigMirror (309 stars, Python): AI agent that auto-generates academic-style charts from your data. A solid example of vertical AI research tooling.
Monthly Trending Cross-Reference
Five repos appeared on both weekly and monthly charts — the strongest "sustained interest" signal:
| Repo | Monthly rank | Monthly +stars | Signal |
|---|---|---|---|
| colbymchenry/codegraph | #1 | +23,688 | Code knowledge graph dominance continues |
| Lum1104/Understand-Anything | #7 | +20,742 | Community validates graph-first approach |
| Imbad0202/academic-research-skills | #6 | +17,780 | Stable demand for academic skills |
| rohitg00/ai-engineering-from-scratch | #14 | +12,957 | Lasting appetite for structured AI engineering learning |
| rohitg00/agentmemory | #5 | +15,782 | Agent persistent memory infrastructure maturing |
Monthly chart entries not in this week's top 15 that are still worth watching: mattpocock/skills (monthly #2, +86,188 stars — highest monthly gain), addyosmani/agent-skills (monthly #21, from Google Chrome engineer Addy Osmani), and multica-ai/andrej-karpathy-skills (monthly #10, +68,832 stars). All three are skills-ecosystem repos, further cementing this month's dominant theme.
This Week's Trend Takeaways
Code knowledge graphs are becoming a new infrastructure layer for AI agents
codegraph and Understand-Anything combined for +35K stars in a single week, both addressing the same engineering bottleneck: AI agents working in large codebases are inefficient because they need extensive tool calls just to understand code structure. Pre-computed knowledge graphs as an "understanding layer" are a logical response. Notably, both repos chose TypeScript — suggesting the primary user base is working in the Node.js/TypeScript ecosystem.
Skills ecosystem moves from individual repos to official organizations
The evolution this week was fast: weekly chart included academic-research-skills (individual), dotnet/skills (official Microsoft), and anthropics/knowledge-work-plugins (official Anthropic). Monthly chart added mattpocock/skills (prominent TypeScript educator), addyosmani/agent-skills (Google Chrome engineer), and multica-ai/andrej-karpathy-skills. This paradigm is expanding from early adopters into mainstream engineering culture. The next thing to watch: will a standardized skills format or registry emerge?
Major companies are open-sourcing agent and security tooling
perplexityai/bumblebee and MoonshotAI/kimi-code are both official releases from major AI companies on this week's new repo chart. Combined with anthropics/financial-services' continued monthly chart presence, AI companies have clearly shifted from "only open-source models" to "open-source the entire toolchain." For developers this is good news — but it also means competition is accelerating, and open toolchains make it easier for others to replicate core capabilities.
Was this article helpful?
