The coding agent ecosystem is shifting away from monolithic prompt hacks toward formal extension architectures and persistent cross-vendor state. Both Anthropic and Cursor dropped competing plugin directories this week, while independent developers rushed to solve the inevitable vendor lock-in with standalone agent memory layers. Meanwhile, quality filtering was ruthless: of 67 raw discoveries swept up this week, exactly zero met our baseline metadata and quality thresholds, exposing a massive wave of low-effort wrappers diluting the index.
A Rust-based long-term memory engine designed for agent coding CLIs to facilitate state handoffs across competing models. Instead of locking context into ephemeral CLI sessions or vendor-specific state, it externalizes task history and architectural decisions so you can bounce between Claude Code, Cursor, or local models without losing your baseline. It tackles the practical bottleneck of agent amnesia at the systems level rather than through brittle system-prompt re-injection.
Anthropic’s official directory and distribution registry for Claude Code plugins. This formalizes Claude Code’s pivot from an opinionated terminal experiment into an extensible developer platform with managed distribution. If you are building workflows inside Claude Code, this repo is now the canonical reference for capability schemas and integration points.
The plugin specification and official plugin library for Cursor. Released alongside Anthropic’s ecosystem push, this defines how third parties expose custom tools and UI capabilities directly to Cursor’s agent. It marks the start of an extension war over where developer context actually lives: inside the editor or inside the terminal runtime.
A real-time spatial intelligence client that streams live, open-source satellite and geospatial feeds onto a photorealistic 3D globe in the browser. It stitches together distributed public earth observation APIs into an interactive, high-performance visualization dashboard. Beyond the cool factor, it serves as
]]>While zero new repositories met our quality threshold this week out of 67 raw discoveries, existing projects saw massive momentum concentrated around agent toolchains and client-side systems. The clearest pattern: engineers are shifting focus from conversational AI wrappers toward modular agent interfaces, local execution fabrics, and deterministically formatted outputs. Between 2-bit models running on microcontrollers and headless iOS virtualization on Apple Silicon, local and edge runtimes are asserting themselves against bloated cloud-only workflows.
Provides a library of 44 self-contained HTML diagram templates engineered explicitly for coding agents like Claude Code, Cursor, and Copilot. Mermaid and raw ASCII often degrade or render inconsistently when generated by LLMs; giving agents direct access to clean, zero-dependency HTML/CSS ensures technical architecture documentation is verifiable and human-readable on first compile.
A self-hosted, private local search engine written in Go that indexes both your browser history and your local filesystem. Instead of offloading your activity history to third-party sync engines or cloud-based note apps, Hister treats personal search as a local-first, privacy-critical utility with minimal overhead.
A unified database management GUI built with Tauri and Rust, offering broad support across MySQL, PostgreSQL, SQLite, Redis, MongoDB, and DuckDB. Most multi-engine database clients are either clunky Java monoliths or RAM-hungry Electron wrappers; DBX demonstrates that Tauri’s web frontend over a native Rust core is the right architecture for fast, lightweight database inspection.
A command-line utility written in Swift that provisions and boots virtual iOS instances directly via Apple’s native Virtualization.framework. It bypasses the overhead of standard GUI simulator suites
This week’s momentum centered on extreme efficiency: native desktop engines replacing Electron bloat, 2-bit foundation models targeting micro-hardware, and low-level virtualization harnesses. While zero unindexed repos crossed our strict discovery quality threshold this cycle, existing tracked projects saw sharp spikes in attention. The overarching trend is an aggressive push toward local-first utility—developers are demanding lightweight, compiled tooling and deterministic execution rather than heavy cloud-dependent abstractions.
t8y2/dbx (Rust, Momentum: 73.6)
A lightweight, multi-engine database management GUI built with Tauri, supporting everything from MySQL, PostgreSQL, and SQLite to Redis, MongoDB, and DuckDB. It trades the memory tax and slow startup of legacy Electron clients for native performance and a single-binary distribution.
handsomestWei/patent-disclosure-skill (Python, Momentum: 72.3)
A specialized agent skill designed for Chinese patent mining, formal disclosure drafting, and statutory examination responses. It proves that agent workflows are rapidly shifting from broad conversational toys to hyper-specific, legally bounded domain tasks.
asciimoo/hister (Go, Momentum: 65.6)
A local, zero-telemetry search engine that indexes browser history, visited URLs, and local files directly on your machine. It offers a fast full-text search backend without leaking personal browsing telemetry or document indices to the cloud.
Lakr233/vphone-cli (Swift, Momentum: 61.6)
A CLI utility that boots virtualized iPhone instances using Apple’s native Virtualization.framework. It unlocks headless, low-overhead iOS environment provisioning on macOS hosts, streamlining low-level testing and automated mobile CI pipelines.
cactus-compute/needle (Python, Momentum: 60.3)
An ultra-compact automation model running at 2-bit quantization with a footprint of just 8–29 MB. Designed for wearables and edge hardware, it delivers local tool calling, structured JSON extraction, and embeddings without requiring network connectivity.
ai-ml)The ai-ml category continues to dominate velocity across the index with an average momentum of 5.4—nearly quadruple any other tracked
2026 Q4 closed with 2,302 repositories tracked, 2,302 of them discovered inside the window. 1,691 clear the quality bar and 7 reach featured grade, spread across 20 active categories.
| Metric | Value |
|---|---|
| Repositories tracked | 2,302 |
| New this window | 2,302 |
| Above tracking bar | 1,691 |
| Featured grade | 7 |
| Combined stars | 30,265,093 |
| Content pieces | 290 |
| Posts with engagement | 660 |
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
149,229 stars · JavaScript · ai-ml · quality 85
🤗 smolagents: a barebones library for agents that think in code.
28,951 stars · Python · ai-ml · quality 85
Official, Anthropic-managed directory of high quality Claude Code Plugins.
37,026 stars · Python · other · quality 82
Command-line tool that allows searching and downloading app packages (known as ipa files) for iOS, iPadOS, tvOS, and visionOS from the App Store.
10,995 stars · Go · cli-tools · quality 82
Boot a Virtual iPhone via Apple’s Virtualization.framework
13,561 stars · Swift · cli-tools · quality 81
A Python library for automating interaction with websites.
4,888 stars · Python · other · quality 81
Extra blocks for scikit-learn pipelines.
1,407 stars · Python · other · quality 80
| Category | Repos | Avg quality | Avg momentum |
|---|---|---|---|
| Other | 578 | 45.2 | 1.0 |
| AI/ML | 405 | 45.5 | 4.3 |
| Web Frameworks | 290 | 55.4 | 0.3 |
| CLI Tools | 205 | 50.3 | 1.2 |
| Databases | 158 | 49.0 | 1.2 |
| DevOps | 121 | 53.7 | 0.6 |
| Design & UX | 113 | 45.1 | 0.8 |
| Developer Tools | 108 | 46.1 | 2.0 |
| Data Science | 56 | 47.2 | 0.0 |
| APIs | 47 | 49.3 | 0.0 |
Fastest mover: developer-tools at +133.3% (7 discoveries versus 3 in the prior window).
Other is the largest slice of the index (578 repos) and the most likely place for next quarter’s movement. The number to watch is momentum, not raw count: a category can grow purely because the scanner looked there more often.
Language distribution across the index: Python (418), TypeScript (369), JavaScript (296), Go (288), Rust (212).
]]>The IoT and hardware category remains a niche segment within the open-source ecosystem, accounting for just 0.5% of the tracked index across 12 active repositories. The category’s average quality score of 40.2 lags behind the ecosystem average of 47.9, reflecting a landscape heavily reliant on static curation lists rather than highly active codebase development. With a language mix spanning C, Python, Shell, and CoffeeScript, the tooling is fragmented, and momentum has been entirely flat across the board over the last 30 days. This indicates a period of quiet maintenance and documentation curation rather than rapid engineering breakthroughs.
The monitoring category remains a quiet, highly fragmented niche in the open-source landscape, accounting for just 0.6% of the overall tracked index. With an average quality score of 46.8, the category slightly underperforms the broader ecosystem average of 47.9. Momentum across almost all 13 tracked repositories is completely flat this month, indicating a period of maintenance rather than active, rapid expansion.
Activity in this category is exceptionally thin, with no standard monitoring projects showing active acceleration or high momentum. The only project demonstrating real growth is AprilNEA/OpenLogi (44.2 momentum), which lies on the boundary of hardware utility and local device monitoring rather than cloud-native observability. The flat momentum across the rest of the index suggests that traditional infrastructure monitoring tools are currently in a stable, quiet maintenance phase.
The monitoring landscape is splitting into two distinct directions. Traditional system and network monitoring (such as uptime trackers and alert consolidators) has largely matured; these projects are stable, but they show zero momentum. Meanwhile, the only real energy and novelty in the category is clustering around specialized runtime environments—specifically local hardware control and the emerging field of AI agent observability. As AI tools and LLMs become core components of the software stack, the definition of “observability” is shifting away from generic host metrics toward tracing agent decision-making loops and API token budgets.
]]>The game development category represents a minor fraction of the tracked open-source ecosystem, accounting for just 0.7% of the total index. The overall health of the category is low, with an average quality score of 34.9, lagging significantly behind the ecosystem average of 47.9. Out of the 17 tracked repositories, 6 are new to this window, indicating some fresh entries, but overall developer momentum is flat across almost all established projects. Furthermore, the dataset suffers from high classification noise, grouping general-purpose business and AI tools alongside actual game engines and titles.
github.com/idavidov13/agentic-playwright)
* github.com/tianchangNorth/pocket-mocker`
Let’s do a quick read-through to ensure flow and constraints are met perfectly.
The education category remains a small niche in the open-source landscape, accounting for just 1.0% of the tracked index with 22 active repositories (7 of which were newly tracked this month). The average quality score of 41.0 lags behind the ecosystem baseline of 47.9, reflecting a mix of highly targeted learning repositories alongside fragmented utility projects. Structurally, the language distribution is led by Python and Clojure (4 repositories each), highlighting a split focus between mainstream machine learning education and niche functional programming toolkits.
The desktop applications category remains a small niche in our tracking, representing just 1.0% of the overall index with 23 tracked repositories. The average quality score of 39.3 sits noticeably below the ecosystem average of 47.9, indicating a high volume of early-stage, experimental, or low-activity projects. In terms of technology, the stack is highly fragmented but shows a clear leaning toward web-adjacent technologies, led by TypeScript (5) and Swift (4), with Tauri (Rust/JS/TS) emerging as a common architectural choice for cross-platform delivery.
This is a mature, highly popular cross-platform GUI client for Clash, built on Tauri and Rust. It stands out for providing a modern, resource-efficient interface across Windows, macOS, and Linux, serving as a prime example of using Tauri to replace heavy, memory-intensive Electron-based desktop clients.
A local-first, Spotlight-style launcher built in Rust that brings semantic search to local files, images, videos, and GitHub stars. By running its search and vectorization fully on-device, it provides a privacy-respecting, low-latency alternative to cloud-dependent search utilities.
A TypeScript-based local control center designed for organizing Codex projects, engineering tasks, and Model Context Protocol (MCP) configurations. It represents a growing class of developer-focused desktop hubs that consolidate local workflows, Git changes, and AI assistant integrations.