This week: audits that disprove their own findings, agents that ask first, and giant models on a desktop

Fastest-growing this period

Counts come from GitHub's official stargazers/history endpoint: the stars gained on each day of the window. Project briefs and the reading notes below are editorial additions.

1. 1

   cloudflare/security-audit-skill : Turn your coding agent into a security auditor that cross-checks its own findings

   JavaScript

   Installs as a skill that makes your coding agent audit a codebase for vulnerabilities. It maps the architecture and trust boundaries first, sends isolated hunter agents after the uncovered areas, then gives every candidate to a fresh verifier that tries to disprove it, so a finding only reaches the report with a complete source trace.

   +15,066
2. 2

   alibaba/open-code-review : Put AI review comments on the lines that need attention

   Go

   Reviews code changes with an AI model and places comments on the relevant lines. File selection and comment positioning are handled by ordinary program logic, useful when you want review suggestions that are easier to locate and check in a large diff.

   +14,857
3. 3

   browser-use/jev-ultrafast : A browser agent that picks each click and typed word in one model call

   Python

   Give it one natural-language goal and it renders the page as a numbered table of controls. TypeSafe's Jev picks the operation and target element in one request, and a small LLM writes text only when a field needs filling, so no site scripts or screenshots drive the loop. The README's measured demo finds flights from Zürich to London in about seven seconds.

   +13,127
4. 4

   hypit-ai/hypit : Let your coding agent turn one reference video into a hundred variants

   TypeScript

   A skill that gives coding agents a language for making video: drop in a reference clip and the agent rebuilds it as a complete workflow covering footage, captions, B-roll and effects, timed to words instead of seconds. You can also start from templates or a plain description, and a workflow can compile captions and motion graphics without calling a generation model.

   +11,797
5. 5

   eternity4719/HowToLiveBetter : Life advice that logs what you spend, what you gain and how solid the evidence is

   HTML

   A Chinese guide of practical suggestions on health, money, scams and legal traps, from chronic disease and first aid to rent disputes and workplace compensation. Each entry records what you spend, what you get back and how strong the evidence is, citing only journal papers and official documents, with an online search page for quick lookups.

   +9,440
6. 6

   deepseek-ai/deepseek-harness : Explore DeepSeek's building blocks for an extensible agent

   TypeScript

   DeepSeek's agent runtime built around replaceable plugins, with a web interface for running it. Useful for developers exploring how to compose an agent's capabilities; it is a developer preview, so expect interfaces to change as you experiment.

   +8,906
7. 7

   bilawalsidhu/gods-eye-view : Follow flights and ships around an interactive 3D Earth

   JavaScript

   A browser-based 3D globe that combines public feeds such as aircraft, ships and earthquakes. Following an aircraft across the terrain is an engaging way to explore geographic data; its visual sensor effects should not be confused with actual sensor imagery.

   +7,542
8. 8

   tt-a1i/archify : Show your team exactly what changed in the architecture

   JavaScript

   Turns an agent's structured description of a system into interactive HTML and SVG architecture diagrams. It can compare diagram snapshots, useful for showing a team which components and connections changed between two versions of a design.

   +7,399
9. 9

   affaan-m/ECC : Give your coding agent a plan, tests and a memory

   JavaScript

   A collection of coding-agent workflows for planning, tests, review and remembering useful context. If your assistant jumps straight into edits or forgets earlier decisions, this gives you reusable routines to adapt to your project.

   +6,492
10. 10

    JustVugg/colibri : Run 700-billion-parameter MoE models on hardware you already own

    C

    A pure C inference engine with zero dependencies that streams the experts of giant mixture-of-experts models from disk, treating VRAM, RAM and storage as a single memory hierarchy. It runs models from 744B up to 2.8T parameters on consumer hardware, and a built-in web dashboard shows which expert tiers are active while you chat.

    +6,288
11. 11

    debpalash/VoiceStudio : Try voice design, transcription and dubbing in one workspace

    Python

    A local audio workspace for voice design, transcription, dubbing and longer narrated projects. Useful for trying several speech workflows with your own material in one application, with engine choice and hardware determining the supported languages and output quality.

    +6,114
12. 12

    stablyai/orca : Compare several coding agents in separate workspaces

    TypeScript

    A desktop interface for running terminal coding agents in separate Git worktrees. You can compare attempts at the same task and inspect their changes, useful when juggling several experiments would otherwise mean managing many terminals and directories.

    +6,006
13. 13

    DietrichGebert/ponytail : Stop your AI turning a small fix into a new framework

    JavaScript

    A set of instructions that asks AI coding agents to reuse existing code and question whether a new abstraction is needed. Its rules also explicitly retain validation and error handling, useful when your agent keeps making a small task unnecessarily complicated.

    +5,836
14. 14

    tamaratran/fast-jev-compaction : Compaction that scores each tool call and drops only the stale ones

    TypeScript

    Replaces Claude Code's compaction summary with per-call decisions. The whole conversation goes to TypeSafe's Jev, every tool call and result gets a keep-or-drop score, stale ones are deleted or truncated, and everything kept stays verbatim and in order, so exact file paths and error messages survive. Ships as both a Claude Code plugin and an npm library.

    +5,522
15. 15

    mattpocock/skills : Make your AI ask the right questions before it writes code

    Shell

    Give your coding assistant a skill that asks detailed questions before implementation. Use it to clarify vague requirements before the agent builds the wrong thing.

    +5,351
16. 16

    Tencent/WeKnora : Turn scattered documents into a self-updating wiki with RAG and agents

    Go

    An open-source knowledge framework from Tencent: ask questions over your documents with RAG, let a ReAct agent orchestrate retrieval, MCP tools and sandboxes, or switch to Wiki Mode, where agents distill raw files into a linked markdown knowledge base with revision history and a knowledge graph. Feishu, GitLab and Notion sources sync in, and the stack self-hosts.

    +5,275
17. 17

    latent-spaces/brag : Turn the project you just built into a shareable launch video

    Python

    A skill that packages the project you just finished into a short launch video with music, motion and ready-to-paste share copy. Run one command in your project folder, steer the tone with a plain-language flag, and you get a rendered video plus the plan and composition brief behind it.

    +5,061
18. 18

    ayghri/i-have-adhd : Stop your coding agent from burying the answer

    Python

    Installs as a skill for Claude Code and similar assistants and changes how they reply: the answer comes first, multi-step tasks are numbered, and every reply ends with one concrete next step. Useful when long agent responses make you hunt for the actual fix; its rule file is plain text, so you can tune the behavior yourself.

    +4,818
19. 19

    Tencent/BrowserSkill : Let your AI agent drive the browser you are already logged into

    TypeScript

    Connects coding agents such as Claude Code and Codex to your everyday browser through a CLI plus extension. Agents run in a separate visible window and borrow only the tabs they need, so you keep working while they reuse your existing logins, and when a captcha or login step appears they can hand control back to you and continue after.

    +4,182
20. 20

    NationalSecurityAgency/ghidra : Decompile and dissect binaries with the NSA's open reverse engineering framework

    Java

    The National Security Agency's open software reverse engineering framework. It disassembles and decompiles compiled code on Windows, macOS and Linux, covers a wide range of processor instruction sets and executable formats, and runs both interactively and in automated mode. Analysts extend it with their own scripts and components written in Java or Python.

    +3,893
21. 21

    abue-ammar/tinycast : A native macOS launcher that also runs your Raycast extensions

    Swift

    A small native macOS command palette built with SwiftUI and AppKit, with no Electron and no telemetry, keeping memory under 100 MB. One hotkey opens fuzzy app search, clipboard history, an inline calculator, window tiling, snippets and custom commands, and it runs the Raycast extensions you already have, rendered as native SwiftUI.

    +3,850
22. 22

    addyosmani/agent-skills : Give your coding agent a repeatable engineering routine

    JavaScript

    A set of coding-agent workflows covering specifications, implementation, tests and review. Each skill includes verification steps, useful when you want to give an assistant a repeatable engineering routine instead of improvising a fresh prompt for every task.

    +3,833
23. 23

    deeplethe/utopia : Trace an AI answer back to its source and its timeline

    Rust

    A self-hosted workspace for searching and discussing documents with source citations. It records both when information applied and when it was learned, making it interesting for examining how an answer changes as a document collection is corrected or updated.

    +3,694
24. 24

    google/artemis : Let an AI assistant drive a real Android phone for tests

    Python

    A Google open-source tool that lets AI assistants and test suites operate a real Android phone or emulator from natural-language instructions. It plugs into coding assistants through MCP, so the assistant can drive the device and pull back screenshots and Logcat logs, while a browser view mirrors the screen live.

    +3,617
25. 25

    bojieli/ai-infra-book : A book that derives LLM system design from hardware limits

    Python

    An open manuscript of a Chinese book on AI infrastructure that derives LLM inference and training design from hardware capacity, bandwidth, and model architecture, using order-of-magnitude estimates instead of generalities. The repo carries the full draft, a typeset PDF rebuilt on every update, and companion experiments and calculation tools.

    +3,491

Ranking window: 2026-09-14 to 2026-09-20 UTC.