2026-05-18

Denoise

AI agent 'skills' are the new prompt engineering, with a specific focus on extending the capabilities of Anthropic's Claude.

Observe the explosion of 'skill' libraries trying to standardize agent behavior, while infrastructure projects aim to connect these agents to CLIs and other real-world tools.

2026-05-182026-05-18T15:51:10Zrules v1Healthyrepos 39signals 39

Top 3 changes

  • CloakHQ/CloakBrowser shows a growing arms race between automated browser agents and sophisticated bot detection systems.
  • HKUDS/CLI-Anything represents the push to make existing software 'agent-native,' connecting LLMs to command-line tools.
  • mattpocock/skills leads a wave of repos attempting to codify and share best practices for instructing AI agents, especially Claude.

Strategic insights

#01An entire ecosystem is rapidly forming around Anthropic's Claude, evident in the numerous 'skill' libraries (mattpocock/skills, tech-leads-club/agent-skills) and dedicated tools (ruvnet/ruflo, zilliztech/claude-context). This is open source building on a closed platform.
#02The definition of an 'agent' is expanding beyond code generation. Projects like HKUDS/CLI-Anything (controlling CLIs) and ruvnet/RuView (interpreting WiFi signals) aim to give agents senses and control over diverse, non-code environments.
#03The 'skill' abstraction is an attempt to move beyond ad-hoc prompting. Repos like humanlayer/12-factor-agents and multica-ai/andrej-karpathy-skills are trying to establish engineering principles for reliable agent behavior, creating a shareable knowledge layer.
#04The rise of CloakHQ/CloakBrowser indicates a critical friction point for agent adoption. As agents become more capable at web automation, they run into increasingly sophisticated anti-bot measures, creating demand for stealth tooling.
#05Financial services is a primary target vertical for agent applications. This is visible in both specialized agent frameworks like TauricResearch/TradingAgents and LLM-driven analysis tools like ZhuLinsen/daily_stock_analysis and Fincept-Corporation/FinceptTerminal.

Trend tracking

Today

A massive surge from `ggml-org/llama.cpp` leads a day dominated by new projects, including `ruvnet/RuView` and stock analysis tool `ZhuLinsen/daily_stock_analysis`.

4 Strengthening · 3 Continuous · 8 First seen
Expand list (15)
ProjectYesterday ⭐Today ⭐ΔState
ggml-org/llama.cpp111k🆕First seen
ruvnet/RuView60k🆕First seen
ZhuLinsen/daily_stock_analysis37k🆕First seen
humanlayer/12-factor-agents20k🆕First seen
CloakHQ/CloakBrowser15k🆕First seen
Imbad0202/academic-research-skills11k🆕First seen
supertone-inc/supertonic8.1k🆕First seen
NVlabs/Sana6.4k🆕First seen
tinyhumansai/openhuman12k16k+4351Strengthening
tech-leads-club/agent-skills2.6k3.9k+1279Strengthening
HKUDS/CLI-Anything35k36k+1111Strengthening
microsoft/ai-agents-for-beginners62k63k+1064Continuous
BigBodyCobain/Shadowbroker6.8k7.6k+811Strengthening
plausible/analytics25k26k+668Continuous
K-Dense-AI/scientific-agent-skills24k24k+658Continuous

This week

Educational repositories show persistence, with newcomer `Imbad0202/academic-research-skills` joining the consistently growing `mattpocock/skills` and privacy-focused `CloakHQ/CloakBrowser`.

5 Strengthening · 5 Continuous · 1 Stable · 2 First seen
Expand list (13)
ProjectYesterday ⭐Today ⭐ΔState
Imbad0202/academic-research-skills11k🆕First seen
colbymchenry/codegraph4.4k🆕First seen
mattpocock/skills88k91k+3211Strengthening
CloakHQ/CloakBrowser13k15k+1583Strengthening
rohitg00/agentmemory11k12k+1324Strengthening
anthropics/financial-services24k25k+1051Strengthening
ruvnet/RuView59k60k+752Continuous
yikart/AiToEarn14k15k+620Strengthening
oven-sh/bun92k92k+300Stable
bytedance/UI-TARS-desktop34k35k+251Continuous
millionco/react-doctor9.9k10k+245Continuous
apernet/hysteria21k21k+172Continuous
facebook/pyrefly6.1k6.2k+107Continuous

This month

AI-focused educational repos show long-term entrenchment, with `mattpocock/skills`, `multica-ai/andrej-karpathy-skills`, and `NousResearch/hermes-agent` all demonstrating sustained accumulation.

6 Strengthening · 11 Continuous · 1 Stable
Expand list (18)
ProjectYesterday ⭐Today ⭐ΔState
mattpocock/skills88k91k+3211Strengthening
multica-ai/andrej-karpathy-skills133k136k+2122Continuous
NousResearch/hermes-agent154k156k+1874Continuous
CloakHQ/CloakBrowser13k15k+1583Strengthening
rohitg00/agentmemory11k12k+1324Strengthening
anthropics/financial-services24k25k+1051Strengthening
Anil-matcha/Open-Generative-AI15k16k+847Strengthening
decolua/9router11k12k+681Strengthening
ruvnet/ruflo52k53k+536Continuous
addyosmani/agent-skills43k43k+508Continuous
TauricResearch/TradingAgents76k77k+484Continuous
heygen-com/hyperframes19k19k+411Continuous
Alishahryar1/free-claude-code25k25k+357Continuous
AIDC-AI/Pixelle-Video18k18k+348Continuous
Z4nzu/hackingtool75k75k+221Stable
soxoj/maigret29k29k+219Continuous
Fincept-Corporation/FinceptTerminal21k22k+201Continuous
zilliztech/claude-context11k11k+122Continuous

Categories· 8

Developer Tools & Productivity(9)

This category highlights a split between traditional high-performance tooling like oven-sh/bun and a new class of AI-assisted tools like millionco/react-doctor. The outlier, CloakHQ/CloakBrowser, addresses the practical challenge of bot detection in automated web workflows, a growing concern for agent developers.

  • A modified Chromium browser designed to evade bot detection systems. Its high ranking is a strong signal of an escalating arms race between web automation (often powered by agents) and services trying to block them. It's a pragmatic tool for developers building scrapers and automated testers.

    rising⭐ 15k
  • A fast, all-in-one JavaScript runtime, bundler, and package manager. Its continued popularity underscores the persistent demand in the JS ecosystem for performance improvements and simplified tooling, challenging the established Node.js and Webpack stack.

    new⭐ 92k
  • A financial analysis terminal application. Distinct from agent-based trading, this is a tool for human analysts, indicating a parallel trend of using modern interfaces and data visualization to empower human decision-making in finance, not just automate it.

    cooling⭐ 22k
  • An agent-based tool that identifies and flags suboptimal React code. This represents a new category of developer tooling: AI-powered linters that go beyond simple syntax checks to enforce best practices and performance patterns based on a model's understanding of a framework.

    new⭐ 10k
  • A proxy tool designed to be fast and censorship-resistant. Its trend suggests a continued user need for robust tools to bypass network restrictions and surveillance, a persistent theme in network utility software.

    new⭐ 21k
  • A fast Python type checker from Meta. Its emergence signals that even with the rise of dynamic languages and AI-assisted coding, large-scale software engineering still demands high-performance static analysis tools for code quality and maintainability.

    new⭐ 6.2k
  • A curriculum from Microsoft for learning to build AI agents. This is a top-down educational effort by a major platform player to onboard developers onto their agent-building stack, aiming to standardize practices and attract users to their ecosystem.

    new⭐ 63k
  • An LLM-driven stock analysis system for Chinese and US markets. This is a practical application showing how individuals are combining data feeds, news APIs, and LLMs to create personalized, automated financial analysis dashboards, democratizing a previously institutional capability.

    new⭐ 37k
  • An open-source, privacy-focused alternative to Google Analytics. Its popularity reflects a structural shift in developer and user sentiment towards privacy, creating demand for tools that provide website insights without extensive data collection or third-party cookies.

    new⭐ 26k

Prompt & Skill Libraries(9)

This category is the epicenter of today's trend. It's a gold rush to define and standardize instructions for AI agents, overwhelmingly targeting Anthropic's Claude. The repos range from personal notes collections (mattpocock/skills) to more formal registries (tech-leads-club/agent-skills), signaling a maturing practice.

  • A collection of prompts and techniques for getting better results from Claude, shared by a well-known developer. Its high visibility is turning a personal configuration file into a de facto standard, kicking off the trend of 'skill sharing' for AI agents.

    repeated⭐ 91k
  • A set of prompts and a structured workflow for using AI in academic research. This is a domain-specific application of the 'skills' trend, demonstrating how structured prompting can be tailored to professional workflows like literature review and paper writing.

    rising⭐ 11k
  • A single prompt file for Claude Code based on Andrej Karpathy's commentary on LLM coding. This is an example of knowledge distillation, where insights from a respected expert are codified into a reusable instruction set for an AI agent.

    cooling⭐ 136k
  • A collection of agent skills tailored for scientific and financial analysis. This project, along with others in its category, shows an effort to create specialized, vertical-specific prompt libraries, moving beyond general-purpose coding instructions.

    new⭐ 24k
  • A repository collecting methods to monetize AI usage. This is more of a list than a tool and reflects the speculative hype cycle around AI, where interest is high in finding novel, often low-effort, ways to generate income from the technology.

    new⭐ 15k
  • An official repository from Anthropic, likely containing examples or prompts for financial applications. Its reappearance indicates sustained interest from the model provider in promoting its use within this lucrative vertical, encouraging ecosystem development.

    repeated⭐ 25k
  • An attempt to create a secure and validated registry for AI agent skills. This represents the next logical step in the 'skills' trend: moving from informal sharing of text files to a more structured, trusted platform for distributing agent capabilities.

    new⭐ 3.9k
  • A collection of 'production-grade' skills for coding agents from a prominent Google engineer. Like mattpocock/skills, this leverages an individual's reputation to lend credibility to a set of prompts, further solidifying the trend of expert-led prompt curation.

    cooling⭐ 43k
  • An attempt to define a set of principles for building production-ready LLM applications, analogous to the '12-Factor App' methodology for web services. This signals a push for maturity and engineering discipline in the agent development space.

    new⭐ 20k

On-device & Multimodal AI(5)

This is a diverse category showing multiple facets of AI interaction. It includes on-device execution with supertone-inc/supertonic for TTS, novel sensor inputs with ruvnet/RuView using WiFi, and programmatic video generation with heygen-com/hyperframes. It's less about a single trend and more about expanding AI's I/O capabilities.

  • An automated short-form video generation engine. This project taps into the demand for content creation automation, particularly for social media platforms like TikTok and YouTube Shorts. It combines multiple AI capabilities (scripting, voiceover, visuals) into a single pipeline.

    cooling⭐ 18k
  • An on-device, multilingual text-to-speech (TTS) engine using the ONNX format. The key signal is its on-device nature, which offers privacy and low-latency performance, contrasting with the many cloud-based AI services trending today. It's a practical component for building private AI applications.

    new⭐ 8.1k
  • A project that uses commodity WiFi signals for spatial sensing and presence detection. This is a novel input modality for AI, moving beyond cameras and microphones. It points to a future where agents can perceive and react to the physical environment in non-visual ways.

    rising⭐ 60k
  • A library for rendering video from HTML, designed to be used by AI agents. This provides a structured, declarative way for an agent to generate video content, abstracting away complex video editing APIs. It's an interesting I/O layer for creative AI.

    cooling⭐ 19k
  • An open-source, self-hostable interface for running many different image and video generation models. Its main draw is the lack of content filters and the ability to run a wide variety of models locally, appealing to users seeking unrestricted creative control.

    cooling⭐ 16k

Memory, RAG & Context(3)

These tools address a fundamental limitation of LLMs: finite context windows. All three projects—rohitg00/agentmemory, colbymchenry/codegraph, and zilliztech/claude-context—are attempting to provide persistent memory or indexed codebase access to give coding agents long-term recall and full project awareness.

  • A library for providing persistent memory to AI coding agents. It addresses a key weakness of current LLMs—their lack of long-term memory—by offering a storage and retrieval mechanism, essential for tasks requiring context from previous interactions or large codebases.

    repeated⭐ 12k
  • Provides a pre-indexed code knowledge graph for AI agents to consume. This is a context-optimization technique, aiming to feed agents structured, relevant information about a codebase rather than raw files, potentially improving accuracy while reducing token usage.

    new⭐ 4.4k
  • A tool from the makers of the Milvus vector database, specifically for providing large-scale context to Claude Code. It's a strategic move by a vector DB company to position its technology as a core component of the emerging Claude developer stack.

    cooling⭐ 11k

AI Infra & Protocols(6)

The focus is on making agents useful by connecting them to external systems. HKUDS/CLI-Anything provides a universal CLI interface, while others like TauricResearch/TradingAgents and ruvnet/ruflo offer specialized orchestration frameworks for finance and Claude-based swarms.

  • An open-source project aiming to build a 'personal AI super intelligence.' Its trending status reflects continued interest in creating private, powerful, and locally-run AI agents that can be personalized over time, moving beyond single-shot interactions with public services.

    new⭐ 16k
  • This project aims to make any command-line interface (CLI) tool usable by an AI agent. It's a significant infrastructure play, attempting to bridge the gap between modern LLMs and the vast ecosystem of decades-old, reliable CLI software, unlocking new automation possibilities.

    new⭐ 36k
  • A framework for building multi-agent systems for financial trading. Its presence indicates a strong push to apply sophisticated agentic workflows, involving multiple specialized LLMs, to the high-stakes domain of automated trading and market analysis.

    cooling⭐ 77k
  • An open-source multimodal AI agent stack from ByteDance. Its value lies in providing a reference architecture for connecting large models to user interfaces and agent infrastructure, showing how major tech companies are thinking about building agent-native applications.

    new⭐ 35k
  • An agent orchestration platform specifically for Anthropic's Claude. It is a clear signal of the ecosystem building around Claude, providing enterprise-focused tools for deploying and managing multi-agent systems built on that specific foundation model.

    cooling⭐ 53k
  • A research project focused on creating an AI agent that improves over time through interaction. While a recurring trend, it represents the ongoing academic and open-source effort to solve the problem of continual learning for LLM agents.

    cooling⭐ 156k

ML & GPU Infrastructure(2)

This category features a stable, widely-used inference engine (ggml-org/llama.cpp) alongside a new research model architecture from a major lab (NVlabs/Sana). The presence of Sana, a Diffusion Transformer, points to ongoing architectural evolution in high-resolution image synthesis beyond current U-Net models.

  • A C/C++ library for running LLMs efficiently on commodity hardware. Its consistent presence on the trending list is a testament to its foundational role in the local AI movement, enabling developers and researchers to run powerful models without specialized GPU hardware.

    new⭐ 111k
  • A research paper and implementation from NVIDIA for image synthesis using a Linear Diffusion Transformer. This is significant as it represents a potential architectural shift from the U-Net backbones common in diffusion models, promising more efficient high-resolution generation.

    new⭐ 6.4k

Security & Reverse Engineering(3)

The focus here is on open-source intelligence (OSINT) automation. Tools like soxoj/maigret and BigBodyCobain/Shadowbroker aggregate public data to build profiles on individuals or track global assets. Z4nzu/hackingtool is a common collection of scripts and not a strategic signal.

  • A collection of various hacking scripts and tools. This type of repository trends frequently and is typically a curated list for beginners or script kiddies. It represents noise rather than a specific, novel development in the security space.

    cooling⭐ 75k
  • An open-source intelligence (OSINT) tool that finds a person's accounts across thousands of websites based on a username. Its continued relevance shows the enduring utility of automated OSINT tools for cybersecurity professionals, investigators, and researchers.

    cooling⭐ 29k
  • An open-source intelligence dashboard for tracking a wide range of global assets, from private jets to satellites. It democratizes access to data that was once the domain of intelligence agencies, enabling hobbyists and researchers to monitor global events.

    new⭐ 7.6k

AI Coding Tools & Agents(2)

This category is dominated by wrappers and routers that provide free or optimized access to proprietary models, specifically Claude Code. Both Alishahryar1/free-claude-code and decolua/9router indicate strong developer demand that current official access models or pricing are not fully meeting.

  • A tool that provides free terminal and VSCode access to Claude Code, likely by proxying requests through free tiers or other means. Its popularity is a direct measure of developer demand for the model, which seemingly exceeds the supply of free or affordable official access.

    cooling⭐ 25k
  • A router that connects various AI coding tools to multiple free model providers. It's a meta-tool for arbitrage and resilience, automatically falling back between services to ensure uptime and minimize costs. This reflects a user base trying to work around API limits and fees.

    cooling⭐ 12k

Recent reports

  • 2026-06-03

    generated 2026-06-03T17:05:52Z · rules v1 · repos 45 · signals 45

  • 2026-06-02

    generated 2026-06-02T16:49:13Z · rules v1 · repos 43 · signals 43

  • 2026-06-01

    generated 2026-06-01T17:57:29Z · rules v1 · repos 44 · signals 44

  • 2026-05-31

    generated 2026-05-31T14:03:33Z · rules v1 · repos 44 · signals 44

  • 2026-05-30

    generated 2026-05-30T13:57:40Z · rules v1 · repos 48 · signals 48

  • 2026-05-29

    generated 2026-05-29T16:02:16Z · rules v1 · repos 46 · signals 46

  • 2026-05-28

    generated 2026-05-28T16:19:16Z · rules v1 · repos 48 · signals 48

  • 2026-05-27

    generated 2026-05-27T16:04:14Z · rules v1 · repos 45 · signals 45

  • 2026-05-26

    generated 2026-05-26T16:01:07Z · rules v1 · repos 45 · signals 45

  • 2026-05-25

    generated 2026-05-25T15:24:58Z · rules v1 · repos 45 · signals 45

  • 2026-05-24

    generated 2026-05-24T13:51:52Z · rules v1 · repos 43 · signals 43

  • 2026-05-23

    generated 2026-05-23T13:55:04Z · rules v1 · repos 42 · signals 42

  • 2026-05-22

    generated 2026-05-22T14:56:21Z · rules v1 · repos 40 · signals 40

  • 2026-05-21

    generated 2026-05-21T15:46:27Z · rules v1 · repos 41 · signals 41

  • 2026-05-20

    generated 2026-05-20T15:48:11Z · rules v1 · repos 36 · signals 36

  • 2026-05-19

    generated 2026-05-19T15:42:42Z · rules v1 · repos 37 · signals 37

  • 2026-05-18 · current

    generated 2026-05-18T15:51:10Z · rules v1 · repos 39 · signals 39

  • 2026-05-17

    generated 2026-05-17T13:49:34Z · rules v1 · repos 45 · signals 45

  • 2026-05-16

    generated 2026-05-16T13:49:33Z · rules v1 · repos 34 · signals 34

  • 2026-05-15

    generated 2026-05-15T14:26:57Z · rules v1 · repos 38 · signals 38

  • 2026-05-14

    generated 2026-05-14T14:32:26Z · rules v1 · repos 42 · signals 42

  • 2026-05-13

    generated 2026-05-13T14:53:18Z · rules v1 · repos 44 · signals 44

  • 2026-05-12

    generated 2026-05-12T14:42:52Z · rules v1 · repos 36 · signals 36

  • 2026-05-11 · fallback

    generated 2026-05-11T15:30:34Z · rules v1 · repos 40 · signals 40

  • 2026-05-10

    generated 2026-05-10T13:46:39Z · rules v1 · repos 37 · signals 37

  • 2026-05-09

    generated 2026-05-09T13:43:42Z · rules v1 · repos 40 · signals 40

  • 2026-05-08

    generated 2026-05-08T14:08:16Z · rules v1 · repos 38 · signals 38

  • 2026-05-07

    generated 2026-05-07T14:38:09Z · rules v1 · repos 37 · signals 37

  • 2026-05-06

    generated 2026-05-06T14:34:27Z · rules v1 · repos 36 · signals 36

  • 2026-05-04

    generated 2026-05-04T14:22:31Z · rules v1 · repos 36 · signals 36

  • 2026-05-03

    generated 2026-05-03T13:40:38Z · rules v1 · repos 29 · signals 29

  • 2026-05-02 · fallback

    generated 2026-05-02T13:39:14Z · rules v1 · repos 30 · signals 30

  • 2026-05-01 · fallback

    generated 2026-05-01T13:46:05Z · rules v1 · repos 32 · signals 32

  • 2026-04-30

    generated 2026-04-30T14:19:05Z · rules v1 · repos 36 · signals 36

  • 2026-04-29

    generated 2026-04-29T14:23:30Z · rules v1 · repos 36 · signals 36

  • 2026-04-28

    generated 2026-04-28T14:34:52Z · rules v1 · repos 37 · signals 37

  • 2026-04-27

    generated 2026-04-27T14:20:15Z · rules v1 · repos 38 · signals 38

  • 2026-04-26

    generated 2026-04-26T13:24:10Z · rules v1 · repos 37 · signals 37

  • 2026-04-25

    generated 2026-04-25T13:22:57Z · rules v1 · repos 41 · signals 41

  • 2026-04-24

    generated 2026-04-24T13:58:04Z · rules v1 · repos 40 · signals 40

  • 2026-04-23

    generated 2026-04-23T14:04:35Z · rules v1 · repos 43 · signals 43

  • 2026-04-22

    generated 2026-04-22T14:02:49Z · rules v1 · repos 41 · signals 41

  • 2026-04-21

    generated 2026-04-22T07:58:11Z · rules v1 · repos 40 · signals 40

  • 2026-04-20

    generated 2026-04-21T09:23:28Z · rules v1 · repos 40 · signals 40

  • 2026-04-19

    generated 2026-04-19T08:51:11Z · rules v1 · repos 36 · signals 36

  • 2026-04-18

    generated 2026-04-18T08:45:02Z · rules v1 · repos 39 · signals 39

  • 2026-04-17

    generated 2026-04-17T09:14:47Z · rules v1 · repos 40 · signals 40

  • 2026-04-03

    generated 2026-05-13T03:24:00Z · rules v1 · repos 26 · signals 26

  • 2026-04-02 · degraded

    generated 2026-04-03T19:29:39Z · rules v1 · repos 26 · signals 26

  • 2026-04-01 · fallback

    generated 2026-04-03T19:29:39Z · rules v1 · repos 26 · signals 26