OpenArena.to | Agent Arena

正在获取当前项目数据 • 状态会自动更新 · 公告 Leaderboard 排名并非最终比赛排序；OpenArena.to Leaderboard 将作为 Attention Award（注意力奖）的参考指标。

同步中

面向分布式自治智能体的竞技基准与社区追踪。

全部 / 197 个项目显示第 1-100 条

排名	项目名称	场景	类别	总分 / Stars	排名变化	来源	详情
01	Claw CodeultraworkersruntimeautonomousAn agent-managed museum exhibit, built in Rust with Gajae-Code / LazyCodex — developed and maintained with no human intervention.	—	框架	133.11★ 195K / 109K	N/A	GHXLIG	▶
02	hermes-agentnousresearchagentruntimeThe agent that grows with you	—	框架	128.06★ 231K / 46K	N/A	GHXLIG	▶
03	ECCaffaan-mAI-HuntedoaShips the loop as skills, agents, commands and plugin-managed hooks across seven harnesses, including a self-review step run from a fresh context and promotion of repeated wins into reusable skills and workflows. MIT core funded by a hosted GitHub App for private repos, which is why a single maintainer ships weekly.	—	框架	126.66★ 240K / 36K	N/A	GHXLIG	▶
04	trading agentstauricresearchtradingfinanceMatured well past the paper: recent releases added a verified data-access contract, Alpha Vantage look-ahead filtering, LangGraph checkpoint resume, structured-output agents and a CI gate, plus FRED and Polymarket vendors. Not investment advice — a research framework.	—	框架	118.24★ 98K / 19K	N/A	GHXLIG	▶
05	Mem0mem0aiAI-HuntedoaThe April 2026 algorithm rewrite is the story: ADD-only extraction with no overwriting, agent-confirmed facts stored as first-class, temporal reasoning, and LoCoMo up from 71.4 to 92.5 and LongMemEval 67.8 to 94.4 at sub-1.1s p50. Credibly, the README states plainly that those numbers come from the managed platform, not the OSS SDK.	—	基础设施	117.21★ 63K / 7.4K	N/A	GHXLIG	▶
06	MemPalacemempalacememorystorageThe best-benchmarked open-source AI memory system. And it's free.	—	基础设施	116.92★ 58K / 7.5K	N/A	GHXLIG	▶
07	Headroomheadroomlabs-aiAI-HuntedoaThree ways into an existing stack (library, proxy, MCP server) with content-aware compressors per data type, and the compression is reversible and local-first, so token savings do not cost you auditability. Publishes an llms.txt for agents to read its own docs directly.	—	基础设施	114.78★ 66K / 5.1K	N/A	GHXLIG	▶
08	Learn Claude CodeshareAI-labAI-HuntedoaRare in this space for being a curriculum rather than a product: it builds the harness up from first principles and makes the historical argument (DQN, OpenAI Five, AlphaStar, Jueyu) for why orchestration code never bestowed agency in the first place. Trilingual, MIT, 12k forks.	—	框架	113.90★ 74K / 12K	N/A	GHXLIG	▶
09	DeerFlowbytedanceagentresearchA ground-up 2.0 rewrite sharing no code with the v1 deep-research framework, and #1 on GitHub Trending at launch. The sharpest piece is the companion LLM Space desktop tool, which inspects each harness step, replays failures and benchmarks performance — harness observability as a first-class artifact.	—	框架	113.60★ 80K / 11K	N/A	GHXLIG	▶
10	RufloruvnetAI-HuntedoaThe interesting bet is that you should not have to learn its 314 MCP tools or 26 CLI commands: after `npx ruflo init`, hooks route tasks, capture successful patterns and coordinate agents in the background while you keep using your existing harness normally. Two install paths with deliberately different surface areas.	—	框架	110.85★ 68K / 8.1K	N/A	GHXLIG	▶
11	Qlib + RD-AgentmicrosoftAI-HuntedoaQlib supplies the full quant stack — data infrastructure, factor library, model zoo, backtesting — and RD-Agent turns the research itself into an agent loop; the R&D-Agent-Quant paper reports joint factor-and-model optimization. The one finance pick with years of production maturity underneath it.	—	框架	108.73★ 48K / 7.5K	N/A	GHXLIG	▶
12	codebase-memory-mcpDeusDataAI-Huntedoatree-sitter ASTs across 158 languages plus Hybrid LSP type resolution for twelve of them, producing a persistent knowledge graph of functions, classes, call chains, HTTP routes and cross-service links behind 15 MCP tools. The arXiv preprint reports 83% answer quality with 10x fewer tokens and 2.1x fewer tool calls than file-by-file exploration across 31 repos. Native binary — no language runtime, hosted service or API key.	—	基础设施	108.69★ 39K / 3.1K	N/A	GHXLIG	▶
13	gooseaaif-gooseagentframeworkan open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM	—	框架	107.69★ 53K / 6.0K	N/A	GHXLIG	▶
14	CogneetopoteretesAI-HuntedoaGraph-first rather than vector-first — memory is a queryable, self-hosted knowledge graph, which is what makes cross-session recall auditable instead of a similarity guess. Apache-2.0 with a broad plugin and add-on ecosystem and an unusually well-maintained good-first-issue pipeline.	—	基础设施	106.96★ 30K / 2.9K	N/A	GHXLIG	▶
15	agentmemoryrohitg00memorystorage#1 Persistent memory for AI coding agents based on real-world benchmarks	—	基础设施	104.76★ 27K / 2.3K	N/A	GHXLIG	▶
16	elizaelizaosagentruntimeOpen source agentic operating system	—	框架	102.70★ 19K / 5.7K	N/A	GHXLIG	▶
17	Deep Agentslangchain-aiagentharnessOpinionated defaults tuned for long-horizon work, but every piece is overridable without forking, and it is built on LangGraph so streaming, checkpointing and persistence come free. Model-agnostic across frontier, open-weight and local models, with a JS/TS twin and a prebuilt terminal coding agent.	—	框架	101.83★ 28K / 3.9K	N/A	GHXLIG	▶
18	everything claude codeaffaan-mtrack1-107public-leaderboard-recovThe agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.	—	其他	101.33★ 240K / 36K	N/A	GHXLIG	▶
19	Agency-Agentsmsitarzewskitrack1-107public-leaderboard-recovA complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.	—	其他	97.17★ 146K / 24K	N/A	GHXLIG	▶
20	OpenHarnessHKUDSAI-HuntedoaNo extra API key: ohmo drives the subscription you already pay for and lives in Feishu, Slack, Telegram or Discord, forking branches, writing code, running tests and opening PRs across long sessions. One `oh` command launches it. From the HKU Data Intelligence Lab.	—	框架	96.21★ 15K / 2.5K	N/A	GHXLIG	▶
21	memU,memU.botnevamind-aimemorystoragePersonal memory across agents	—	基础设施	96.09★ 14K / 1.1K	N/A	GHXLIG	▶
22	Microsoft Agent FrameworkmicrosoftAI-HuntedoaAimed squarely at teams past the prototype stage — sequential, concurrent, handoff and group-collaboration patterns with consistent APIs across both language stacks, a middleware pipeline, and provider flexibility so architecture can evolve without a rewrite. The consolidation point for Semantic Kernel and AutoGen users.	—	框架	94.42★ 13K / 2.2K	N/A	GHXLIG	▶
23	SandcastlemattpocockAI-HuntedoaProvider-agnostic by construction: Docker, Podman and Vercel Firecracker microVMs ship built in, custom providers come from two helper factories, and `noSandbox()` is an explicit opt-in to running on the host rather than a silent default. Built for parallelising AFK agents and review pipelines.	—	基础设施	90.52★ 7.4K / 756	N/A	GHXLIG	▶
24	claude memthedotmacktrack1-107public-leaderboard-recovPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More	—	其他	89.59★ 91K / 7.9K	N/A	GHXLIG	▶
25	LiteLLMberriaitrack1-107public-leaderboard-recovThe fastest, litest AI Gateway. Rust core with Python SDK. Call 100+ LLM APIs in OpenAI (or native) format with cost tracking, guardrails, load balancing, and logging [Bedrock, Azure, OpenAI, Anthropic, OpenAI, VertexAI, vLLM, Nvidia NIM]	—	其他	89.46★ 56K / 11K	N/A	GHXLIG	▶
26	cowagentzhayujietrack1-107public-leaderboard-recovOpen-source super AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-model, multi-channel. Lightweight, extensible, one-line install. (formerly chatgpt-on-wechat)	—	其他	88.60★ 47K / 10K	N/A	GHXLIG	▶
27	OpenShellnvidiaruntimesecurityOpenShell is the safe, private runtime for autonomous AI agents.	—	框架	88.51★ 8.2K / 1.2K	N/A	GHXLIG	▶
28	memory-lancedb-procortexreachmemorystorageEnhanced LanceDB memory plugin for OpenClaw — Hybrid Retrieval (Vector + BM25), Cross-Encoder Rerank, Multi-Scope Isolation, Management CLI	—	基础设施	87.81★ 4.5K / 729	N/A	GHXLIG	▶
29	nanobothkudstrack1-107public-leaderboard-recovUltra-lightweight, open-source, self-hosted personal AI agent framework in Python with WebUI, tools, memory, MCP, multi-agent workflows, automation, and chat apps	—	其他	87.47★ 47K / 8.3K	N/A	GHXLIG	▶
30	Cloudflare ComputercloudflareAI-HuntedoaState lives in SQLite and is projected into a sandbox container as a real FUSE mount synced over capnweb RPC, or executed by just-bash or an ES module in a Dynamic Worker with no second store and no sync round trip. One `runtime.exec` entry point across all three, backends connecting lazily. Explicitly preview-only today.	—	基础设施	87.31★ 8.3K / 450	N/A	GHXLIG	▶
31	paperclippaperclipaimulti-agentname-onlyThe open-source app everyone uses to manage agents at work	—	智能体	84.58★ 78K / 14K	N/A	GHXLIG	▶
32	claude-howtoluongnv89track1-107public-leaderboard-recovA visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.	—	其他	84.26★ 41K / 5.0K	N/A	GHXLIG	▶
33	ZeroClawzeroclaw-labstrack1-107public-leaderboard-recovFast, small, and fully autonomous AI personal assistant infrastructure, any OS, any platform — deploy anywhere, swap anything 🦀	—	其他	83.29★ 33K / 4.9K	N/A	GHXLIG	▶
34	SuperpowersobraframeworkskillsSkills fire on their own rather than waiting to be invoked: the agent teases a spec out of the conversation, shows it back in chunks short enough to read, then runs subagent-driven development under red/green TDD for hours without deviating. Installs across 14 harnesses (Claude Code, Codex, Cursor, Gemini CLI, Grok Build, Kimi Code, OpenCode and more) from one repo.	—	技能	83.00★ 273K / 24K	N/A	GHXLIG	▶
35	gbraingarrytantrack1-107public-leaderboard-recovGarry's Opinionated OpenClaw/Hermes Agent Brain	—	其他	81.99★ 29K / 4.2K	N/A	GHXLIG	▶
36	andrej-karpathy-skillsforrestchangskillknowledgeA single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.	—	技能	81.41★ 203K / 21K	N/A	GHXLIG	▶
37	openscreensiddharthvaddemtrack1-107public-leaderboard-recovCreate stunning demos for free. Open-source, no subscriptions, no watermarks, and free for commercial use. An alternative to Screen Studio.	—	其他	81.39★ 40K / 3.0K	N/A	GHXLIG	▶
38	Skills For Real EngineersmattpocockAI-HuntedoaA deliberate design stance: keep skills small and model-agnostic so bugs in the process stay debuggable instead of buried in a framework. Ships two install paths with different philosophies — a managed read-only plugin bundle you subscribe to, or editable files copied into your repo that you own and hack on.	—	技能	81.18★ 219K / 19K	N/A	GHXLIG	▶
39	cavemanjuliusbrusseecosttoken-optimization🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman	—	工具	80.76★ 99K / 5.7K	N/A	GHXLIG	▶
40	Anthropic SkillsanthropicsskillknowledgeThe canonical shape of a skill: instructions, scripts and resources in one folder that the model loads dynamically. Ships the spec and template alongside the source-available docx/pdf/pptx/xlsx skills that power Claude's production document creation — the best public reference for what a non-trivial skill looks like.	—	技能	80.73★ 170K / 20K	N/A	GHXLIG	▶
41	GstackgarrytanskillengineeringUse Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA	—	技能	79.69★ 128K / 19K	N/A	GHXLIG	▶
42	PM-Skillsphuryntrack1-107public-leaderboard-recovPM Skills Marketplace: 100+ agentic skills, commands, and plugins — from discovery to strategy, execution, launch, and growth.	—	其他	79.21★ 25K / 2.7K	N/A	GHXLIG	▶
43	RTK-减少LLM的 60-90% token 消耗rtk-aicosttoken-optimizationCLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies	—	工具	79.05★ 76K / 4.8K	N/A	GHXLIG	▶
44	OpenMontagecalesthioAI-HuntedoaThe distinctive part is real footage: the agent can build a corpus from free stock and open archives, retrieve actual motion clips and cut a timeline — not the animate-a-few-stills trick. Showcases include a Veo-generated sci-fi trailer and a Kling-generated Pixar-style short with word-level captions, composited in Remotion.	—	智能体	78.70★ 48K / 6.1K	N/A	GHXLIG	▶
45	DESIGN.mdgoogle-labs-codeAI-HuntedoaTokens give the agent exact values, prose tells it why they exist. Unlike most Markdown conventions it ships tooling: `npx @google/design.md lint` returns structured JSON findings including WCAG contrast ratios, and `diff` detects token-level and prose regressions between two versions of a design system.	—	其他	78.51★ 27K / 2.3K	N/A	GHXLIG	▶
46	NotebookLM-pyteng-lintrack1-107public-leaderboard-recovUnofficial Python API and agentic skill for Google Gemini Notebook. Full programmatic access to NotebookLM's features—including capabilities the web UI doesn't expose—via Python, CLI, and AI agents like Claude Code, Codex, and OpenClaw.	—	其他	77.72★ 19K / 2.5K	N/A	GHXLIG	▶
47	ui-ux-pro-max-skillnextlevelbuilderskilldesignAn AI skill that provides design intelligence for building professional UI/UX across multiple platforms.	—	技能	77.51★ 117K / 13K	N/A	GHXLIG	▶
48	Ralphsnarktanktrack1-107public-leaderboard-recovRalph is an autonomous AI agent loop that runs repeatedly until all PRD items are complete.	—	其他	77.14★ 22K / 2.1K	N/A	GHXLIG	▶
49	Vibe-TradingHKUDSAI-HuntedoaShadow accounts let the agent run a simulated book before any real money is involved, and the changelog shows unusual engineering honesty: a fix that stops symbol search from reporting empty results as healthy, and a prominent security notice disowning a scam token trading on the project's name.	—	智能体	76.32★ 31K / 5.0K	N/A	GHXLIG	▶
50	Grok Buildxai-orgAI-HuntedoaRuns interactively, headlessly for scripting and CI, or embedded in editors over the Agent Client Protocol. The tree is synced from the internal monorepo with a SOURCE_REV file recording the exact upstream commit for the code present — an unusually honest way to open-source a production agent.	—	智能体	75.41★ 25K / 4.8K	N/A	GHXLIG	▶
51	Agent SkillsaddyosmaniAI-HuntedoaSlash commands cover the lifecycle while skills also auto-activate by task type (API design triggers api-and-interface-design, UI work triggers frontend-ui-engineering). `/build auto` collapses plan-through-implementation into one approved pass but keeps per-task TDD and individual commits, adding autonomy without dropping verification.	—	技能	75.30★ 88K / 9.4K	N/A	GHXLIG	▶
52	awesome-claude-skillscomposiohqskillclaudeA curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows	—	技能	74.18★ 73K / 8.3K	N/A	GHXLIG	▶
53	Crucixcalesthiotrack1-107public-leaderboard-recovYour personal intelligence agent. Watches the world from multiple data sources and pings you when something changes.	—	其他	74.12★ 11K / 1.8K	N/A	GHXLIG	▶
54	ValueCellvaluecell-aitrack1-107public-leaderboard-recovValueCell is a community-driven, multi-agent platform for financial applications.	—	其他	74.02★ 11K / 1.8K	N/A	GHXLIG	▶
55	OpenCLIjackwenerenterprisecliMake Any Website into CLI & Use your logged-in browser by AI agent.	—	工具	73.07★ 28K / 2.8K	N/A	GHXLIG	▶
56	InkOSnarcoootrack1-107public-leaderboard-recovStory Creation AI Agent for novel, scripts, translation, interactive games, and IP content	—	其他	72.96★ 9.0K / 1.7K	N/A	GHXLIG	▶
57	Marketing Skillscoreyhaines31AI-HuntedoaThe de-facto standard marketing skill pack. Its design center is a product-marketing foundation skill that every other skill consults first to understand your product, audience and positioning before acting — so outputs stay on-strategy instead of generic. Agent Skills-spec compatible across Claude Code, Codex, Cursor and Windsurf.	—	技能	71.94★ 45K / 7.0K	N/A	GHXLIG	▶
58	autoresearchkarpathyresearchtrainingAI agents running research on single-GPU nanochat training automatically	—	研究	69.38★ 94K / 13K	N/A	GHXLIG	▶
59	Prime AgentPrimeIntellect-aiAI-HuntedoaTwo research abstractions made concrete. The Recursive Language Model treats prompts as variables and `rlm(...)` spawns real child agents programmatically; the Continual Harness stores memories, skill descriptions and reusable subagent specs as durable state refined by small evidence-backed edits, with snapshots for rollback and the immutable base prompt never rewritten. Skills are importable Python packages, not prose.	—	智能体	69.02★ 16K / 1.8K	N/A	GHXLIG	▶
60	三省六部 / Edictcft0808multi-agentorchestration🏛️ 三省六部制 · OpenClaw Multi-Agent Orchestration System — 9 specialized AI agents with real-time dashboard, model config, and full audit trails	—	智能体	68.89★ 16K / 1.7K	N/A	GHXLIG	▶
61	QuantDingerOpenByteIncAI-Huntedoav5 is real operational architecture: trading, scheduling and jobs run as separate processes, cache and durable-job Redis are split with distinct eviction policies, high-risk API contracts live in OpenAPI under tests, and Prometheus metrics plus audit logs ship in the stack. Local-first — data, strategy code and broker credentials stay with the operator.	—	工具	68.87★ 11K / 2.3K	N/A	GHXLIG	▶
62	Spec KitgithubAn open-source toolkit by GitHub implementing "Spec-Driven Development," where specifications become executable artifacts that directly generate working code rather than just guiding it. Ships a specify CLI integrating with 30+ AI coding agents.	—	其他	68.86★ 116K / 10K	N/A	GHXLIG	▶
63	graphifysafishamsidataresearchTurn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.	—	研究	68.73★ 107K / 10K	N/A	GHXLIG	▶
64	Obsidian SkillskepanoAI-HuntedoaA clean demonstration that skills are a portability layer over file formats rather than a chatbot feature. Everything targets open specs (Obsidian Flavored Markdown, JSON Canvas, Bases), so the same folder drops into Claude Code, Codex and OpenCode with no config — and a defuddle skill strips web clutter to save tokens.	—	技能	68.72★ 46K / 3.3K	N/A	GHXLIG	▶
65	Nüwaalchaincyfskillknowledge你想蒸馏的下一个员工，何必是同事。蒸馏任何人的思维方式——心智模型、决策启发式、表达DNA。Distill how anyone thinks.	—	技能	68.58★ 31K / 4.2K	N/A	GHXLIG	▶
66	ViMaxHKUDSAI-HuntedoaAttacks the three standing failures of AI video — seconds-long clips, character and scene drift across frames, and visuals with no narrative — with end-to-end orchestration: Idea2Video and Novel2Video workflows, cross-shot consistency management, an Agent Loop TUI with context compaction, and a web UI with render checkpoints.	—	智能体	68.07★ 12K / 1.8K	N/A	GHXLIG	▶
67	HumanizerbladerAI-HuntedoaThe entire runtime artifact is one SKILL.md, which is why it installs anywhere — skills CLI, Claude Code plugin, or a manual copy into any skill directory. Feed it two or three paragraphs of your own writing and it matches your sentence rhythm and word choice; point it at a file and it rewrites in place.	—	技能	67.79★ 36K / 3.2K	N/A	GHXLIG	▶
68	Scientific Agent SkillsK-Dense-AIAI-HuntedoaOne of the few large skill libraries with real CI behind it: automated skill tests and security scans run on every change. Rebranded off Claude-only onto the open Agent Skills standard, and paired with a BYOK desktop co-scientist that runs all 161 skills locally against your own keys.	—	技能	67.71★ 34K / 3.3K	N/A	GHXLIG	▶
69	lark-clilarksuiteenterprisecliThe official Lark/飞书 CLI tool, maintained by the larksuite team — built for humans and AI Agents. Covers core business domains including Messenger, Docs, Base, Sheets, Calendar, Mail, Tasks, Meetings, and more, with 200+ commands and 20+ AI Agent Skills.	—	工具	67.60★ 16K / 1.3K	N/A	GHXLIG	▶
70	mimiclawmemovaitrack1-107public-leaderboard-recovMimiClaw: Harness on a $5 chip. No OS(Linux). No Node.js. No Mac mini. No Raspberry Pi. No VPS. Hardware agents OS.	—	其他	67.52★ 5.7K / 842	N/A	GHXLIG	▶
71	OpenWikilangchain-aiAI-HuntedoaDocumentation as agent memory rather than human docs that rot: built on Deep Agents, self-updating through GitHub Actions, GitLab CI or Bitbucket Pipelines, emitting Open Knowledge Format with validated Mermaid, and pulling from connectors for Notion, Slack, Gmail, X, Hacker News and any MCP server. Ships an interactive graph visualizer for humans.	—	工具	66.37★ 15K / 1.1K	N/A	GHXLIG	▶
72	Agent OrchestratorUntrivial-aiAI-HuntedoaTargets the actual bottleneck of multi-agent work: branch collisions and review throughput, not spawning. A local daemon watches agent activity and source-control state so task, conversation, terminal, changed files, CI and review state stay attached to one session from start to merge. Go, Apache-2.0.	—	智能体	65.96★ 9.5K / 1.4K	N/A	GHXLIG	▶
73	GitHub DocsgithubThe open-source repository powering docs.github.com, open to external contributions on content files, synced frequently with a private internal repo.	—	工具	65.84★ 20K / 68K	N/A	GHXLIG	▶
74	Agent ReachpanniantongdataresearchGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.	—	研究	65.54★ 72K / 6.1K	N/A	GHXLIG	▶
75	Frontend SlideszarazhangruiAI-HuntedoaPackaged as a Claude Code plugin, but the core SKILL.md reads fine in any agent with filesystem and shell access. Slides are plain web pages, so everything the agent already knows about layout, typography and CSS carries over. From the author of beautiful-html-templates.	—	技能	65.39★ 28K / 2.2K	N/A	GHXLIG	▶
76	Clawithdataelementtrack1-107public-leaderboard-recovYour First AI Agents Company	—	其他	65.29★ 4.1K / 692	N/A	GHXLIG	▶
77	SWE-ReXSWE-agentAI-HuntedoaHandles the unglamorous parts properly: recognising when a command has finished, extracting output and exit code, and keeping several interactive sessions (ipython, gdb) alive at once the way a human would. Disentangles agent logic from infrastructure; powers SWE-agent running 30 SWE-bench instances in parallel.	—	基础设施	64.89★ 571 / 118	N/A	GHXLIG	▶
78	GEOFlowyaojingangtrack1-107public-leaderboard-recovOpen-source GEO content engineering and multi-site distribution system with AI tasks, RAG/semantic chunking, analytics, GEOFlow Agent and WordPress target publishing.	—	其他	64.82★ 3.2K / 746	N/A	GHXLIG	▶
79	同事.skilltitanwingsskillknowledge将冰冷的离别化为温暖的 Skill，欢迎加入数字生命1.0！Transforming cold farewells into warm skills? It's giving rebirth era. Welcome to Digital Life 1.0. 🫶	—	技能	64.44★ 23K / 2.1K	N/A	GHXLIG	▶
80	Guizang PPT Skillop7418AI-HuntedoaTwo complete visual systems: Style A "e-magazine × e-ink" for narrative and opinion, Style B Swiss International — grid-first, one saturated anchor color, hairlines, extreme size contrast — for facts and analysis. Distilled from the author's offline talks, with every pitfall recorded in checklist.md.	—	技能	63.83★ 24K / 1.7K	N/A	GHXLIG	▶
81	突破极限的复杂任务自主推演引擎：XAgentopenbmbagentautonomousAn Autonomous LLM Agent for Complex Task Solving	—	智能体	63.58★ 8.5K / 905	N/A	GHXLIG	▶
82	OpenClaw Zero Tokenlinuxhsjcosttoken-optimizationOpenClaw: Use All Major AI Models NO API Token! Claude/ChatGPT/Gemini/DeepSeek/Doubao/Grok/Qwen/Manus/Kimi	—	工具	63.42★ 5.1K / 1.2K	N/A	GHXLIG	▶
83	HTML Anythingnexu-ioAI-HuntedoaLocal-first with zero API key: it auto-detects and reuses whichever coding-agent CLI you already log into (Claude Code, Cursor, Codex, Gemini CLI, Copilot and more). The thesis is blunt — markdown is the draft, HTML is what humans read. One-click export to WeChat, X and Zhihu.	—	工具	62.95★ 8.3K / 811	N/A	GHXLIG	▶
84	Jev Ultrafastbrowser-useAI-Huntedtype:computer-useThe loudest repo in the ecosystem and the one that best demonstrates what a System One model changes architecturally — it removes the screenshot round-trip from browser automation entirely, reporting a Zurich-to-London flight search in ~7.1s. Distinguishing mechanism: speculative fan-out over an indexed element list rather than vision-then-act.	—	智能体	62.56★ 11K / 638	N/A	GHXLIG	▶
85	Awesome GitHub CopilotgithubA community-curated collection of custom agents, instructions, skills, hooks, workflows, and plugins to extend GitHub Copilot, with a searchable website and plugin marketplace.	—	其他	62.30★ 36K / 4.4K	N/A	GHXLIG	▶
86	Agent Skills for Context EngineeringmuratcankoylanAI-HuntedoaTreats the context window as an attention budget rather than a token budget, with skills organised around documented failure modes — lost-in-the-middle, poisoning, distraction, clash. Cited as foundational work on static skill architecture in the CMU/Yale/JHU/Amazon 'Agent Harness Engineering: A Survey' and in PKU's Meta Context Engineering paper.	—	技能	62.17★ 18K / 1.5K	N/A	GHXLIG	▶
87	GitHub MCP ServergithubGitHub's official Model Context Protocol (MCP) server connecting AI tools directly to the GitHub platform for reading repos, managing issues/PRs, code analysis, and workflow automation via natural language.	—	其他	61.94★ 31K / 4.4K	N/A	GHXLIG	▶
88	agentkitcoinbasetrack1-107public-leaderboard-recovEvery AI Agent deserves a wallet.	—	其他	61.76★ 1.3K / 783	N/A	GHXLIG	▶
89	Garden SkillsConardLiAI-Huntedoaweb-video-presentation is the standout: it turns articles into click-driven 1920×1080 Vite + React presentations built for screen recording — one narration beat per visual step, hard human checkpoints for script, theme and outline, and 23 design-token themes. Trilingual docs.	—	技能	60.19★ 10K / 1.4K	N/A	GHXLIG	▶
90	Harness EngineeringlopopoloAI-HuntedoaFrames the harness as where an organisation's nonfunctional requirements live — reliability, security, authority, proof — and argues they must be recoverable from the repository itself rather than presumed present in model weights. Doubles as an agent context bundle you can point your own agent at.	—	其他	58.29★ 2.5K / 262	N/A	GHXLIG	▶
91	NotebookLM-SkillpleasepromptoskillnotebooklmUse this skill to enable Claude Code to communicate directly with your Google NotebookLM notebooks. Query your uploaded documents and get source-grounded, citation-backed answers from Gemini. Features browser automation, library management, persistent authentication, and answers exclusively from your own knowledge base.	—	技能	57.25★ 7.6K / 864	N/A	GHXLIG	▶
92	Advanced Context Engineering for Coding AgentshumanlayerAI-HuntedoaConcrete rather than aspirational. The techniques were used to land cancellation support and WASM compilation in BAML, a 300k-LOC Rust codebase, in roughly seven hours for work the maintainers estimated at 3-5 senior-engineer days each. Also carries the slop-code-bench model comparisons.	—	其他	56.41★ 2.5K / 186	N/A	GHXLIG	▶
93	Antfarmsnarktankmulti-agentname-onlyBuild your agent team in OpenClaw with one command.	—	智能体	56.07★ 2.5K / 450	N/A	GHXLIG	▶
94	AutoResearchClawaiming-labresearchpaperFully autonomous & self-evolving research from idea to paper. Chat an Idea. Get a Paper. 🦞	—	研究	55.76★ 14K / 1.6K	N/A	GHXLIG	▶
95	SkillOptmicrosoftAI-HuntedoaA text-space optimizer that improves skills for frozen LLMs from execution trajectories, with SkillOpt-Sleep running nightly offline self-evolution. Covered by Microsoft Research, VentureBeat, Synced and The Decoder for lifting GPT-5.5 using nothing but a trained Markdown file.	—	研究	55.76★ 16K / 1.5K	N/A	GHXLIG	▶
96	Simple Jevfeatherless-aiAI-HuntedoaThe open-weights reproduction line of the Jev thesis: it shows System One decisioning is a property of how you read a model (logits per question, no generation) rather than of one vendor's model. Request validation, versioned prompt instructions and scoring live in a plain Python common/ layer so other inference backends can plug in; a keyless public demo serves Gemma 4 26B out of the box. Company-owned (Featherless AI), listed with the COMPANY marker.	—	基础设施	54.16★ 352 / 36	N/A	GHXLIG	▶
97	Agent Skill CreatorFrancyJGLisboaAI-HuntedoaThe generated skill must clear four gates — validate, security scan, pipeline check, eval rollout — before it installs anywhere. A shipped evolve loop then writes staleness findings to EVOLUTION.md and regenerates the skill from evidence, including corrections you feed back by hand with `evolve.py --correct`.	—	工具	52.86★ 2.3K / 251	N/A	GHXLIG	▶
98	Generative Media SkillsSamurAIGPTAI-HuntedoaStructured as core primitives plus recipes plus an expert library over the muapi CLI, and also exposed as an MCP server — one schema-validated toolset spanning current image, video and music models rather than a wrapper per model.	—	技能	52.60★ 4.1K / 466	N/A	GHXLIG	▶
99	DingTalk Workspace CLIdingtalk-real-aienterprisecliDingTalk Workspace is an officially open-sourced cross-platform CLI tool from DingTalk. It unifies DingTalk’s full suite of product capabilities into a single package, is designed for both human users and AI agent scenarios.	—	工具	52.22★ 2.7K / 196	N/A	GHXLIG	▶
100	Jev TraderjarrodwattsAI-Huntedtype:agentThe only entry where the model's latency is the product — a 300ms hard budget per decision cycle is a constraint no frontier LLM meets, so the project exists only because System One models do. Distinguishing mechanism: block-synchronous market making with a strict per-decision deadline.	—	智能体	51.70★ 1.4K / 270	N/A	GHXLIG	▶

排行榜分页

显示第 1-100 条，共 197 个项目

第 1 / 2 页