An open-source toolkit packing 28 research skills into a structured workflow that runs from literature review to paper submission, with 9 verifiable knowledge bases and broad compatibility with mainstream AI coding clients. Built for researchers who want to automate the grunt work of the academic pipeline.
Explore the radar’s memory
Every pick we’ve ever shipped, plotted as a living constellation. Picks cluster by topic; threads between clusters reveal cross-topic connections. Zoom in to expand a cluster into its picks — drag, filter, click through to the brief.
// Browse
All picks
Archive results
-
Light0305/Light-skills
-
shy3130/tickflow-stock-panel
A self-hosted quantitative workbench for Chinese A-share markets — stock screening, monitoring, and backtesting in one panel. Runs without operational overhead, plugs into third-party data sources like Tushare, and scales with different subscription tiers.
-
tigicion/dao-code
A terminal coding agent built on DeepSeek V4 that runs long tasks without losing context, self-verifies its own output, and stays cheap. For solo founders and indie hackers who want agentic coding in the terminal without paying Claude or GPT-4 rates.
-
eddyzzl/marvis-risk-agent
An agentic framework aimed at credit risk workflows — model development, validation, feature engineering, and strategy in one place. Niche enough that most builders will pass, but if you are shipping fintech or lending infrastructure, this is worth twenty minutes.
-
duckbugio/flock
An open-source autonomous AI dev-team bot that coordinates multiple agents to plan, code, and ship features with minimal human steering. Worth a look if you're running a tiny team and want to delegate chunks of the backlog to something that does more than just autocomplete.
-
dzcmemory-web/bazi-ziwei-skill
An agent skill that runs traditional Chinese astrology calculations — Bazi and Ziwei Doushu — using deterministic algorithm code rather than LLM guessing, with three analysis modes and one-click generation of an ink-style HTML chart poster.
-
hoolulu/deep-research
One command triggers a multi-step research pipeline that returns a structured, professional report in up to 19 languages in about ten minutes. Aimed at founders or analysts who need a defensible research artifact fast, not just a chatbot summary.
-
omnigent-ai/omnigent
An open-source meta-harness that lets you orchestrate Claude Code, Codex, Cursor, and custom agents through a single interface. Swap the underlying agent without rewriting orchestration logic, enforce policies across all of them, and coordinate from any device in real time.
-
Eynzof/Hermes-CN-Desktop
A desktop wrapper for the Hermes Agent, built with Tauri, TypeScript, and Rust. Windows-first. Packages an isolated Hermes Agent core for local use. Of interest if you specifically want a native desktop shell around Hermes rather than a web-based interface.
-
helloianneo/ian-xiaohei-illustrations
A skill for generating 16:9 white-background hand-drawn Chinese-style illustrations with sparse red, orange, and blue annotations. Aimed at creators building content in the grotesque or whimsical Chinese illustration style. Narrow use case, but fills a specific aesthetic gap.
-
tastyeffectco/sandboxes
Spins up self-hosted dev sandboxes with live preview URLs in a single command, no Kubernetes required. Built for coding agents and SaaS factories that need throwaway environments fast. If you are wiring up an AI coding loop and do not want to pay per-sandbox cloud fees, this is worth your Saturday.
-
nexu-io/html-video
Converts HTML, CSS, and data into real MP4 files locally, no per-render fees and no cloud dependency. Comes with 21 templates, pluggable render engines, and an AI soundtrack option. Built for coding agents that need programmatic video output without a SaaS bill attached.
-
husu/loom
An AI agent that writes API documentation via natural-language conversation, then bundles a viewer and mock server alongside the docs it generates. Aimed at backend teams who hate maintaining documentation as a separate chore from shipping code.
-
lessweb/deepcode-cli
A terminal coding assistant built specifically for DeepSeek models, with controls for reasoning depth and agent-style skill execution. If you are already running DeepSeek locally or via API and want CLI-native code help rather than a browser IDE, this is the open-source project to bench this weekend.
-
op7418/guizang-ppt-skill
An agent skill that generates HTML slide decks with editorial magazine layouts, Swiss grid aesthetics, image prompts, and social covers — plus a WebGL presentation runtime. Aimed at anyone who wants outputs that do not look like a GPT wrote them.
-
browser-use/browser-harness
A self-healing browser harness that lets LLMs complete arbitrary tasks in a real browser, recovering automatically when the page structure breaks or changes. Built for teams tired of brittle selectors dying every time a vendor updates their UI.
-
OTT-Cybersecurity-LLC/lyrie-ai
Self-described autonomous AI cybersecurity agent from a small security firm. The repo is light on substance and heavy on claims. Cybersecurity is a crowded space with better-validated tools; monitor rather than adopt today.
-
brycewang-stanford/Awesome-Agent-Skills-for-Empirical-Research
A library of 23,000-plus agent skills aimed at social science empirical research, maintained alongside a tool that claims to generate a reproducible paper in 20 minutes. Niche but deep — relevant if you are building research-automation tooling or academic workflow products.
-
oritera/Cairn
A general-purpose state-space search engine that the authors validated on autonomous penetration testing. The abstraction is interesting — if your problem can be framed as a search over states, Cairn wants to be the engine. Early stage, but worth a look for AI engineers building agentic planning systems.
-
worldwonderer/oh-story-claudecode
A Claude Code skill pack for Chinese web-novel writers, covering the full pipeline from outline scanning and story deconstruction through drafting, AI-ness removal, and cover generation. Niche by design — useful if you are writing long-form serialized fiction in Chinese and already live in Claude Code.
-
thClaws/thClaws
A self-hosted agent harness written in Rust, supporting multiple model providers. Aimed at teams that want to run agentic workflows on their own infrastructure without a SaaS middleman. Worth a look if local sovereignty over your agent execution environment is a hard requirement.
-
agents-io/PokeClaw
A fully local AI agent that physically operates an Android phone — tapping, swiping, filling forms — using Gemma 4 with no cloud calls, no API key, and no backend required. Anyone building Android automation, accessibility tools, or offline AI agents will want to see how far this no-dependency approach actually goes.
-
Xiangyue-Zhang/auto-deep-researcher-24x7
An autonomous agent that runs deep learning experiments unattended using a Leader-Worker architecture with constant-size memory. Aimed at ML researchers who want to queue experiments overnight without babysitting GPU jobs. Early-stage but the zero-cost monitoring angle is worth a glance.
-
ciembor/agent-rules-books
If your AI coding agent keeps producing structurally sloppy code, this collection of AGENTS.md rule files gives it a backbone by encoding principles from Clean Code, DDD, and Clean Architecture directly into its instructions. Engineers running Cursor or Claude Code on serious codebases will want to see how far opinionated rules can actually shape agent output.
-
AMAP-ML/SkillClaw
A framework for letting AI agents evolve skills collectively via an agentic evolver loop. Research-tier project from AMAP-ML. Relevant to ML researchers building multi-agent systems, less so to anyone shipping a product this quarter.