开源 AI agent 框架，已安装并运行

赋予 coding agent 品味的 agent 技能：艺术指导、参考图片和浏览器检查，而不是通用的 AI 页面

测试发现：install.sh symlinks all 14 skills into ~/.codex/skills, skill.sh resolves skill names to SKILL.md paths, draw.mjs deals art directions from a library of 247 curated sites, lint.mjs scans source files for banned patterns, and all 5 unit tests in tests/skill_test.sh pass. 3 分钟。

一个开源的长时程 SuperAgent harness，用于研究、编码和创作。借助沙箱、记忆、工具、技能、子代理和消息网关，它可以处理不同级别的任务，这些任务可能需要几分钟到几小时。

测试发现：Backend Gateway boots and serves health checks on port 8003 (HTTP 200), 1642 backend tests pass, 160 blocking-IO tests pass, 24811 frontend unit tests pass, all dependencies installed. 62 分钟。

在你的机器上运行的求职搜索。基于 Claude Code 构建的 AI 求职申请框架：评估职位信息、定制简历、撰写求职信、准备面试。Fork 它并拥有它。

测试发现：Python test suite (321 unittest tests) and all 6 Bun portal CLI test suites (231 tests) pass. LinkedIn search CLI makes live queries and returns results. 4 Python analysis tools (security_guards, framework_version, lint_skills, verify_pdf) run successfully. 8 分钟。

停止租用你的智能，用 AnythingLLM 拥有它，获得强大的本地优先 agent 体验所需的一切。

测试发现：The AnythingLLM monorepo installs and builds cleanly; the server boots on port 3001 responding to API calls; the collector boots on port 8888; all 1,119 unit tests across server (725) and collector (394) pass without errors. 20 分钟。

你的 coding agent 所依赖的运行时

测试发现：herdr 0.8.2 builds, runs its server, passes all Rust unit tests, Python maintenance tests, Bun integration tests, and plugin marketplace tests with no failures. 19 分钟。

开源个人人工智能助手与 Agent Harness。规划任务，运行工具和技能，通过记忆和知识自我进化。多 Agent、多模型、多通道。轻量级、可扩展、一行安装。

测试发现：CowAgent 2.1.9 installs, builds, all 1949 tests pass, web console binds on 127.0.0.1 and answers /api/health with 200 OK, and the CLI reports version 2.1.9. 17 分钟。

团队优先的 Multi-agent 编排，用于 Claude Code

测试发现：Oh-My-ClaudeCode v5.2.0 CLI tool builds and responds to version/help/info commands; 1,094+ tests pass with 3 env-constrained failures (Node 18 engine gap + shallow git). 31 分钟。

使用 AI coding agents 通过一条命令克隆任何网站

测试发现：Next.js 16.3.0 project builds, typechecks, and serves HTTP 200 on the root route via npm run dev. 4 分钟。

将任何技术书籍 PDF 转化为 Claude Code 技能，可在工作中随时学习、参考和使用。

测试发现：The book-to-skill package is installed as an editable pip package, passes ruff lint, all 627 tests pass, skill validation passes, and the CLI extracts text from documents producing structured JSON metadata and clean text output. 3 分钟。

无限免费 AI 编程。通过 40+ 提供商将 Claude Code、Codex、Cursor、Cline、Copilot、Antigravity 连接到免费的 Claude/GPT/Gemini。自动回退，RTK -40% tokens，永不触及限制。

测试发现：Root npm install completed, npm run build succeeded, the standalone Next.js app serves the login page (200), models API (200 with JSON model list), and version endpoint (200), and the test suite runs 2562 passing tests. 42 分钟。

开源 AI 编排框架，用于构建上下文工程化的生产就绪 LLM 应用。设计具有对检索、路由、记忆和生成显式控制的模块化流水线和 agent 工作流。专为可扩展 agent、RAG、多模态应用、语义搜索和对话系统而构建。

测试发现：Haystack 3.3.0-rc0 installed with Hatch, all 6569 unit tests pass, package builds to sdist and wheel. 4 分钟。

将参考图像中的对象重建为纯代码、程序化、质量门控、动画就绪的 Three.js 模型。Token 高效的图像到 3D。

测试发现：All 1416 forge test suite tests pass. forge/ state.py, next.py, report.py, stage2_spec/validate_sculpt_spec.py, stage3_build/generate_threejs_factory.py all import and run cleanly from Python 3.12 stdlib. 7 分钟。

面向机器人和 agent 的 AI 搜索引擎与 MCP 服务器。单个 Rust 二进制文件即可索引代码、文档、日志、JSONL/CSV 和 PDF，用于 BM25 全文、向量和混合搜索、无需 grep 的代码搜索、agent 记忆与 RAG。兼容 Elasticsearch 的 API；Corpus Hub...

测试发现：XERJ v1.0.0-rc.77 built from source in 2m06s (quick profile), serving ES-compat API on :9200 with autoindex and MCP server working; test suites pass except 2 IPv6 bind failures specific to this container. 36 分钟。

开源 Agent 操作系统

测试发现：OpenFang 0.6.9 compiles, passes 2699 tests, and the daemon starts successfully, booting 30 agents and serving the API on port 4200 with a health endpoint returning HTTP 200. 28 分钟。

随性编码 非严肃开发的从入门到门入的工具箱

测试发现：All repository quality gates pass except check-research-raw which requires gh CLI auth. Node v22 and Python venv with all dependencies are installed from /tmp. Submodules initialized. Markdown lint and all Python-based checks (links, details, structure, directory coverage, metadata, AI citations, external resources... 6 分钟。

面向 AI agent 的计算机使用 CLI

测试发现：npm-installed and source-built agent-desktop 1.1.0 both answer version with ok:true, its 4311 Rust tests pass with 0 failures, and against a real GTK3 app on Xvfb+openbox with the real at-spi2-registryd, snapshot, find, get, click, and screenshot all work, where clicking the Go button changed the button's accessible... 62 分钟。

一套用代码为 Claude Code 制作的电影风格工具箱：66 种现成风格，也可以制作你自己的风格。每种风格都是一个风格提示词，加上一部完全由 Claude Opus 5.5 用代码制作的演示影片，不使用视频模型。作者：Lemomo。 · 一套用代码做片的风格工具箱：66 种现成风格，也能做出你自己的

测试发现：With Node 20.20.2, the core, voice, and music installs exit 0, and the real pipeline produced a 6.000-second 1920x1080 24 fps film with 144 H.264 frames and stereo AAC audio that fully decoded with exit 0. 7 分钟。

面向长视界 agents 的增量引擎，如果你喜欢，请给它加星标。

测试发现：The Python package builds and installs from source, the CLI runs and prints version 999.0.0, 1095 Python tests pass (core engine, ops, resources, connectors that don't need external services), and 144 Rust tests pass across the core, utils, and SDK crates. 12 分钟。

实用的模式、入门模板和 CLI 工具，用于 AI coding agents 的循环工程。设计用于提示和编排 agent 的系统（灵感来自 Addy Osmani 和 Boris Cherny）。包含 loop-audit 、 loop-init 、 loop-cost 。

测试发现：Both documented CLIs build and run from a clean copy, 252 validation-gate tests pass with 0 failures, and the repository audits at 100/100 L2. 10 分钟。

一个更智能的自托管 AI 助手：多用户、多 agent。

测试发现：Python dependencies installed and resolved, 3401 unit tests pass, `octop --version` prints v1.0.2b5, `octop run` starts uvicorn listening on the configured port, and the health endpoint ('/api/health') returns 200. 16 分钟。

为团队中的每个 coding agent 提供一份共享记忆。Claude Code、Cursor 和 Codex 会获得你代码和团队过往决策的地图，从而从你停下的地方继续。

测试发现：The graft CLI installs, builds, and its full test suite passes (1491/1492, 1 deliberate skip) with both previously-failing shim-resolution and lock-release tests green, and the CLI builds/parses/queries a scratch repo and completes an LLM deep pass against a local mock endpoint. 87 分钟。

团队中的每个 coding agent 共享一个记忆。Claude Code、Cursor 和 Codex 获得你的代码和团队过往决策的地图，从而从你停下的地方继续。

测试发现：npm install completes, TypeScript compiles, the CLI builds/checks/queries a structural code graph on a real repo without any keys, and all 1220 tests run with 0 failures. 10 分钟。

一个与 LLM 无关的移动 Agent，可使用自然语言命令自动化你的移动设备

测试发现：Mobilerun 0.6.19 is installed in a Python 3.12 venv. All 979 tests pass (0 failures). The CLI responds to --version and --help. The package imports cleanly. Anthropic SDK is pinned to 1.17.0 for MessageParam compatibility. 45 分钟。

让 AI agents 浏览 web。一个用于基于浏览器的 AI agents 的自主工具包。

测试发现：Bun 1.4.2 installed, all 42 npm dependencies resolved, TypeScript builds pass for all 3 packages, 364/364 core tests pass, CLI prints help/version correctly, and a headless Playwright browser opens and navigates to https://example.com/. 8 分钟。

面向 Agents 的项目管理技能系统，使用 GitHub Issues 和 Git worktrees 实现并行 agent 执行。

测试发现：CCPM skill installed at /home/runner/.codex/skills/ccpm, .claude/ directory initialized with all subdirectories, gh 2.64.0 available via ~/.local/bin/gh, and all 14 local workflow scripts execute with correct output against real data. GitHub sync commands blocked by missing interactive auth. 4 分钟。

Stream 的 Open Vision Agents。使用任何模型或视频提供商快速构建语音和视觉 agent。借助 Stream 的边缘网络实现超低延迟。

测试发现：The vision-agents core framework plus anthropic, gemini, openai, deepgram, elevenlabs, getstream, and smart_turn plugins are installed; 733 unit tests pass, the CLI binary responds, and all selected plugin imports resolve correctly. 42 分钟。

面向 AI agent 和应用的开源浏览器 API。Steel Browser 是开箱即用的浏览器沙箱，让你能够自动化操作网页，无需担心基础设施。

测试发现：With a user-local Node 22 and Chrome for Testing plus DISABLE_CHROME_SANDBOX=true, the dev-mode API runs on port 3000, /v1/health returns 200 with live Chrome, and sessions/scrape/screenshot/pdf plus external CDP client connections all worked with real results. 11 分钟。

由最先进模型驱动的 AI 套件，提供先进的 AI/AGI 功能。包括 AI 角色、AGI 功能、世界级的 Beam 多模型聊天、文生图、语音、响应流式输出、代码高亮与执行、PDF 导入、开发者预设，以及更多功能。可部署在本地或云端。

测试发现：All npm commands (install, test, build, tscheck, lint) pass cleanly on Node.js v22.14.0. The test suite runs 25 subtests: 4 pass (including a live OpenRouter model listing returning 388 models), 21 skip due to absent API keys, 0 fail. 9 分钟。

AI Agent Governance Toolkit：策略执行、零信任身份、执行沙箱和可靠性工程，用于自主AI agent。覆盖 OWASP Agentic Top 10 的 10/10。

测试发现：All core Python packages install from source: agent-governance-toolkit-core, agent-governance-toolkit-cli, agent-control-specification (Rust wheel), agent-governance-toolkit-compliance. The agent-os test suite passes 2993/2993 selected tests, the ACS SDK test suite passes 306/306 selected tests, and the quickstart... 18 分钟。

构建分布式、生产级、长期运行的 agents。

测试发现：AgentScope Java 2.0.4-SNAPSHOT builds and all 3082 test cases across core, harness, and extension modules pass with zero failures. 79 分钟。

用于定义 AI agent 的框架无关、git 原生标准

测试发现：OpenGAP v0.5.0 builds, all 51 unit tests pass, CLI works for init/validate/info/export commands, and validates example agents successfully. 5 分钟。

可移植的 AI agent 编排，带有机械协议执行。186 个 agent，零运行时依赖。

测试发现：Harmonist pack integrates and enforces the agent protocol pipeline on Python 3.12, all 373 test assertions passed across 22 shell/Python test suites covering integration, hooks, upgrade, memory, telemetry, supply-chain, regression, and cross-platform paths; the sole gap is lsof(1) missing from the container, blocking... 6 分钟。

Waku Waku! Waku Agent 是一个本地优先的 AI agent harness，你真正拥有它，包括循环、记忆、评估，全部以代码形式构建，并随着增长保持可读性。

测试发现：Waku 0.1.8 installed and verified: all 1714 offline deterministic tests pass, the dashboard serves HTTP 200 on localhost:7777, the CLI responds to waku --help, the loop runs end-to-end with a real LLM provider (OpenRouter free tier), and lint passes cleanly. 7 分钟。

智能体 LLM 漏洞扫描器 / AI 红队工具包

测试发现：Agentic Security LLM vulnerability scanner is fully installed, all 390 unit and integration tests pass, the FastAPI server starts and responds on /health (HTTP 200) and /v1/self-probe (returns mock OpenAI-style completion). 19 分钟。

一个用于构建生产级 agent 系统的 Go 框架，支持图工作流、工具、记忆、A2A、AG-UI、MCP、评估和可观测性。

测试发现：The tRPC-Agent-Go root module (trpc.group/trpc-go/trpc-agent-go) builds and all 235 test suites pass, the test/ module passes with Go 1.24.4 auto-downloaded, and the examples/ module builds successfully, all from a fresh container with Go installed from the upstream tarball. 14 分钟。

面向 Elixir 的自主 agent 框架。专为分布式、自主行为和动态工作流而构建。

测试发现：Elixir 1.18.5 + Erlang/OTP 25 installed under ~/local_erlang and ~/elixir; mix deps.get, mix compile, and mix test all succeed with 2470 passing tests and 0 failures. 16 分钟。

Bub it. 构建它。一个微小的 agent 运行时，可通过插件组合。

测试发现：Bub installed from source, builds with uv, all 579 tests pass with zero failures, the CLI accepts all commands, and the framework makes live agent calls through the default OpenRouter free model returning correct text responses. 3 分钟。

[GenAI 应用开发框架] 快速轻松构建 GenAI 应用。使用结构化数据和链式调用语法在代码中轻松与 GenAI agent 交互。使用事件驱动流 *TriggerFlow* 管理复杂的 GenAI 工作逻辑。无需重写应用代码即可切换至任何模型。

测试发现：The agently 4.1.4.8 package installs, imports, and creates agents via its fluent API. 3,776 of 3,826 tests pass. 9 failures are pre-existing characterization snapshot mismatches (expected on different package versions), and 1 requires a Docker daemon. 45 分钟。

构建 agents 的标准，更好

测试发现：The better-agents CLI builds, typechecks, and passes 15 unit tests; the init command successfully generates a complete agent project scaffold including src/, prompts/, tests/scenarios/, AGENTS.md, .env, and .mcp.json. 5 分钟。

面向编码 agent 及更多领域的终极 harness，整合你所有的 agent 和机器，形成一个命令中心。从代码开始，然后追随你的好奇心跨领域构建：CAD、电路、机器人、游戏和音乐。

测试发现：The CLI monorepo package installs, typechecks, builds, and passes all 5741 tests on Ubuntu 24.04 with Node 22.23.3, with sqlite3 and lsof made available in the container environment. 48 分钟。

一个 pod 用于你的团队和你各自运行的 agents：你自己的 Claude Code，或一键启动的托管 agent。开源（Apache-2.0），可自托管。

测试发现：Installed deps and built both workspaces, fixed the two real test-bug failures (expired fixture date, missing WebCrypto global), and verified the backend server live on localhost:5000: /api/health 200 healthy with real mongod connected, register 201, login 200 with JWT, while frontend's 962 Jest tests pass and full... 79 分钟。

每次运行默认持久、可重放且可恢复的 agent 框架。

测试发现：Chidori v3.8.1 binary installed at ~/.chidori/bin/chidori; runs TypeScript agents directly via its embedded JS engine, serves agents over HTTP on localhost, replays runs deterministically from journaled call logs with zero LLM calls, scaffolds agent projects from templates, and both the TypeScript and Python SDKs... 8 分钟。

LightAgent：轻量级 Python 框架，用于 OpenAI 兼容的 agent，支持工具、记忆、护栏、追踪、生命周期钩子、多 agent 协作和工作流。

测试发现：LightAgent v0.10.2 installed in a Python 3.12 venv via pip install -e .; 287 unit tests pass covering memory, tools, tracing, guardrails, hooks, evaluation, review, flows, connectors, capabilities, skills, sessions, and runtime hardening; safe_expression and SharedMemoryPool verified working offline. 2 分钟。

开源自动研究，由自主 coding agent 驱动。运行 Claude Code、OpenCode 和 Codex，支持评分、共享知识和多 agent 进化。已收录于 COLM 2026。

测试发现：CORAL CLI works with no errors on a standard dev install. 827 tests pass. Grading, runs catalog, worktree provisioning, and web dashboard all verified working end-to-end. 21 分钟。

ReachAI 企业级智能体开发平台：快速、安全完成已有业务系统智能化改造，让 AI 在 OA、ERP、CRM 等原系统中查数据、填表单、办业务。

测试发现：Backend builds cleanly and all 2246 unit/integration tests pass across 9 Maven modules. Frontend builds with npm run build and 453 of 454 vitest tests pass across 10 test suites (1 pre-existing timezone-dependent dashboard test fails in UTC). 34 分钟。

构建 100% 在设备上运行的 AI agents，在 Qualcomm NPU 上延迟低于 100 毫秒，零云依赖。

测试发现：OAK v0.3.0 OSS build compiles and passes 6/6 tests + 5/5 eval suites on gcc 13.3, no errors. 4 分钟。

智能体智能的基础层

测试发现：Promptise installs in a Python 3.12 venv, the CLI prints version 0.1.dev1, all 3493 selected tests pass, and the package imports without error. 8 分钟。

使 agent 高效且自我进化的语义 agent 内核

测试发现：Three binaries (semantix, semantix-agent, semantix-gateway) build and the full test suite passes with 0 failures for all 148 Go packages including race-detected runs. 17 分钟。

最自主的 AI agent 框架：在 GitHub Actions 上无人值守运行，具备自愈技能，驱动 Claude Code、Grok、Codex 等。无审批循环。一次配置，永久无忧。

测试发现：npm test runs clean: 205 dashboard unit tests and 7 webhook replay-guard tests all pass on Node 18.19.1 after fixing the test script glob and adding 'type':'module' to the webhook package.json. 4 分钟。

长时间运行的嵌入式 agent

测试发现：The repo compiles, all 25 tests pass, and the crabtalkd daemon starts and accepts CLI commands (crabtalk ps, version, info) over a Unix domain socket with no third-party credentials required. 10 分钟。

开源自托管的 AI agent 运行时和多 agent 框架，用于自主 agent 群，提供 Agent 记忆、MCP 工具、调度、委派以及 23+ 个 LLM 提供商（Claude、GPT、Gemini、OpenRouter、Ollama），是 Claude Code 和 LangChain 的实用替代方案。

测试发现：SwarmClaw installs, builds, passes all 938 tests, and runs: the built and dev servers both answer 200 on /api/healthz, the web app serves HTML on /, the CLI answers --help and version 1.9.40, and the Electron desktop app launches on the virtual display and loads its UI. 9 分钟。

面向 AI agents 的 LLM 监督持久记忆：基于图的回忆、跨会话知识、单一二进制。兼容 DeepSeek Harness、Claude Code、OpenClaw 及任何 agent 运行时。

测试发现：All 19 deterministic Go test packages pass, the mnemon binary builds and runs, and the installed binary at $GOBIN/mnemon successfully executes memory operations (remember, recall, status) on a real SQLite-backed store. 5 分钟。

TypeScript AI agent 框架：认知记忆、动态人格、涌现行为与自构建设计

测试发现：AgentOS 0.10.19 builds from source with pnpm install, TypeScript compiles to dist/, and 2100+ vitest tests pass across config, memory, API, runtime, orchestration, RAG, e2e, and cognition modules on Node.js 18.19.1 with better-sqlite3 and hnswlib-node native addons rebuilt manually. 54 分钟。

Daydreams 是一套用于构建商业 agent 的工具。

测试发现：pnpm install succeeds, all 13 packages build, 157/157 core tests pass, core module imports cleanly with 20+ exported symbols. 6 分钟。

本书围绕 DeerFlow 2.0，从理论到源码，系统讲解如何进行二次开发。（精校）

测试发现：The VitePress documentation site for the DeerFlow 二次开发 book installs, builds (0 errors, 18 chapter pages), and serves via vitepress dev on port 5173. 2 分钟。

Forall (∀) 是 Astrio 的一个 coding harness，帮助开发者通过生成规范驱动的代码以及机器可验证的证明来构建正确的软件。

测试发现：Rust workspace builds with cargo, all 44 unit/integration tests pass, and the npm MCP bridge correctly proxies MCP tool requests between stdio and Streamable HTTP. 5 分钟。

AutoHarness：面向 AI Agent 的自动化测试框架工程

测试发现：The AutoHarness CLI and library install and build cleanly in a project venv, the governance pipeline runs for real (validation, allowed/denied tool calls, mode switching, wrap, export, claude-code hook install), and the full test suite passes 958/958; the only mocks needed are LLM credentials for the AgentLoop. 5 分钟。

Harness 导向的 agent 系统框架，用于生产级 LLM agent 应用

测试发现：Hypha builds and runs: the API server serves 200 on /api/v1/status/page, /api/v1/status, /api/v1/openapi.json and completes governed tool executions with verified hashes, and the unit (305), package (2420), integration (44), and native production-topology acceptance (6) tests all pass against locally stood-up MongoDB... 36 分钟。

生产级 Go SDK，用于构建具有长期记忆、知识检索和语音功能的 AI agent，可作为库、守护进程或实时流水线运行。

测试发现：All 7 Go workspace modules pass vet and tests, and the forge demo binary prints its command usage. 6 分钟。

支持文件系统和编程式技能。

测试发现：pydantic-ai-skills works against the current pydantic-ai-harness line: the package installs, all 351 tests pass, and a real agent run loads a bundled skill, reads a reference file, and executes its script, with Git and S3 registries syncing libraries harness accepts. 49 分钟。

面向 Claude Code 的多 agent 团队操作系统。108 个 MCP 工具，40+ 个 agent 模板，10 个生命周期钩子，7 个流水线工作流。持久化团队、结构化会议、任务墙、实时 React 仪表盘。不使用 LangChain/AutoGen，纯粹的 CC 原生集成。

测试发现：AI Team OS v1.14.0 API answers 200 on /api/health, serves 187 REST endpoints and 116 MCP tools. Codex adapter installed with 7/7 hook files, MCP via stdio verified. SQLite database at ~/.claude/data/ai-team-os/aiteam.db holds 47 tables with working project/task CRUD and cross-request persistence. Unit tests pass... 44 分钟。

面向 Rust 的 agent 循环：从 7 种 LLM 协议流式传输，运行工具，循环直到完成。从笔记本电脑到 Cloudflare Worker。

测试发现：yoagent v0.23.0 compiles and its full test suite passes with zero failures on Rust 1.99.1 using only crate-bundled dependencies and MockProvider-based tests. 13 分钟。

为你的 agent 提供合适的 IDE 和操作系统。编码 agent 的感觉运动皮层（OpenCode + Pi），属于 CortexKit 的一部分：符号感知编辑、语义搜索、代码健康、快速 grep/glob、bash 压缩、后台任务、PTY。

测试发现：AFT v0.58.0 builds from source with Rust and TypeScript. The Rust binary (aft) receives JSON-over-stdio commands and responds to version queries. All 3957 Rust unit tests pass. TypeScript plugin build produces working dist bundles. All Pi plugin tests pass (839). OpenCode plugin tests: 1566 pass, 3 are... 55 分钟。

用于构建你的角色机器人的 Event-stream AI Agent 框架

测试发现：pnpm install completes, pnpm test runs 4908 tests all passing, pnpm run typecheck passes cleanly, pnpm start launches the bot with a web console responding on port 7788 with no auth required. 11 分钟。

本地优先、受治理的 Python AI agent 运行时：可嵌入你的应用，或作为 CLI 或 ACP 服务器运行。内置权限、MCP、记忆和审计回放。

测试发现：Agentao 0.5.7.dev0 installs, builds, and passes its entire 6336-test suite inside the container with mock credentials; real LLM chat requires a genuine API key. 6 分钟。

Libra：面向 Agentic RL 后训练的高效资源管理

测试发现：Full install succeeded, 303/303 test suite tests pass, 6 skipped, and 2 of 3 broken example scripts now run end-to-end on CPU-only container with no GPU required. 28 分钟。

面向自主 AI agent 的组织即代码。受大脑启发的记忆，能够增长、整合和遗忘。多模型（Claude/Codex/Gemini/Cursor/Ollama）。

测试发现：17092 unit tests pass; CLI prints full help; demo server binds port 18501 with Uvicorn; all 5 pre-existing test failures fixed via IPv6 loopback detection, zstd Python fallback, and Node 18 ESM compatibility. 44 分钟。

2026 年 AI Agent 框架、工具、平台和资源的精选列表，这一年 agent 走向主流。

测试发现：All Python maintenance scripts and the 10-test unittest suite pass cleanly; the network link check over 1210 URLs exited 0 with 0 confirmed-dead links, the only non-200 results being 75 access-blocked and 5 transport-error URLs that the repository explicitly classifies as unverified rather than broken. 2 分钟。

开源的 WhatsApp AI agent SDK。自带模型、自带号码、自带数据。本地优先，无需强制云，没有厂商锁定。

测试发现：The full npm workspace builds (tsc), passes typecheck (tsc --noEmit), and passes all 1,202 tests across 9 packages with 0 failures after upgrading to Node 22 and adding the missing rolldown optional native binding. 2 分钟。

技能宪法：管理所有 agent 平台上技能调用的元规则

测试发现：Skills Constitution v2.31.0 installs and all 142 regression tests pass. The skill tree (764 standalone + 326 plugin skills) loads, the pre-hook injects context correctly, the constitution-gate blocks/protects as designed, and the workbuddy/cursor/prompt install modes copy files and rebuild the tree without error. 5 分钟。

面向 AI coding agents 的治理框架。它通过五个步骤的工作流（plan、build、review、test、ship）运行它们，任何步骤没有证据都不算完成。通过 AGENTS.md 为 Claude Code、Codex、Cursor、Copilot 和 Antigravity...

测试发现：Full test suite (810 tests) passes, validate.sh reports all PASS, and credential-scan demo runs correctly. 9 分钟。

一个能学习、适应并完成任何任务的伴侣 AI agent。

测试发现：From a clean checkout the standard install steps succeed on this container after user-level Rust and Node 22 installs; the gateway boots, serves the console and authenticated API, and the built Hypermid daemon answers real memory-record create and list requests. 47 分钟。

受 Pi 启发的 Rust 原生、库优先的 coding-agent 运行时和终端 CLI，具有可组合的 LLM 提供方、工具、会话和 plugin ABI

测试发现：Full Rust workspace compiles and all 3176 tests pass; the minimal example runs offline using the faux provider and produces the expected agent transcript; the CLI binary reports version 0.16.0. 13 分钟。

DSPy 的 Go 实现

测试发现：DSPy-Go library and CLI build successfully, all 35 test packages pass, CLI responds to list/describe/version/help commands. 9 分钟。

用于发布就绪短视频包的趋势转视频 Agent 框架

测试发现：Trend2Video Pro installed, all 11 tests pass, CLI generation produces valid MP4 with full publish package (title, description, hashtags, subtitles, thumbnail, metadata, quality report), API responds on /health and /api/score endpoints, benchmark runs, and SQLite database stores topics and generation history. 6 分钟。

Trend-to-Video Agent Framework：用于生成可发布短视频包

测试发现：A fresh install runs all 13 tests, the API returns 200 on its documented endpoints, the Streamlit UI generates a package, and the CLI produces a decodable 30-second 1080x1920 MP4 with 48 valid subtitle cues. 14 分钟。

面向 Claude Code、Codex、Cursor、OpenCode、GitHub Copilot、Google Antigravity 和 Pi 的多 harness 智能体插件市场

测试发现：uv sync installs all deps for both projects; pytest suite passes 619/621; all 6 harness generators emit artifacts; codex doctor reports 20/20 checks ok; structural validation passes across 6 harnesses with strict mode. 11 分钟。

开源 AI coding agent，专为大型项目和真实世界任务而设计。

测试发现：Plandex server starts on port 8099 with PostgreSQL 16.4 and LiteLLM proxy on port 4000; server tests pass; CLI builds and authenticates against local server; full CRUD API for projects, plans, branches, settings is operational. 18 分钟。

让你的 AI agent 像房间里最懒的资深开发一样思考。最好的代码就是你从未写过的代码。

测试发现：All 110 tests pass: the root project's behavioral, plugin, correctness, and hardware-awareness tests (84), the Pi extension's mode/injection tests (23), and the MCP server's instruction-building tests (3) all succeed. 3 分钟。

开源、自托管的 OpenAI Dots 和 Grok Bot 替代品。专为不被反机器人系统检测而设计。

测试发现：npm ci installed 428 packages, the CLI bundle and web client built, npm run typecheck passed, npx vitest run gave 2103 passed/3 skipped and go test gave 120 passed after I fixed the secret-input backspace bug and the execEnv DISPLAY bug, and the control plane served HTTP 200 on /api/health, /api/dots and the web UI... 22 分钟。

数字生命卡兹克开源的 AI 技能合集 | Agent 技能：leader（帮你定义目标），neat-freak 洁癖，hv-analysis，khazix-writer 和更多：Claude Code，Codex 和 40+ agents

测试发现：AIHOT v2.0.0 installed and its live endpoints returned 200, neat-freak's structural suite passed 11 behavior and 21 trigger cases, the PDF converter produced a readable 2-page Chinese PDF, and the storage report server served 200 while rejecting an out-of-allowlist delete with 403; Linux disk scanning returned... 5 分钟。

一个可靠的 coding agent，用于处理复杂的软件工程任务。

测试发现：Reasonix builds as a single Go binary (bin/reasonix), answers --help with full usage including CLI/TUI, web, serve, ACP, and bot subcommands, passes all 147+ test packages with 0 failures, and produces valid JSON diagnostics via doctor --json. 19 分钟。

自托管的 TikTok、Douyin 与 Bilibili 数据采集器和无水印视频下载器，提供异步 REST API、面向 AI agent 的 MCP 服务器、CLI 和 Web 控制台，并支持对已下载视频进行离线语音转文字。具备自愈身份池和 PostgreSQL 归档，一条 docker compose up...

测试发现：All 2960 tests pass (2450 unit + 510 integration), the CLI outputs version 5.0.1 and lists 10 commands, the API server starts and serves healthz/readyz/swagger/console root with 200, the web console SPA is built, and both the PostgreSQL+TimescaleDB and Redis backends and the Go downloader sidecar run. 48 分钟。

Omnigent 是一个开源的 AI agent 框架和 meta-harness：编排 Claude Code、Codex、Cursor、Pi 和自定义 agent；无需重写即可切换 harness，执行策略和沙箱机制，并可从任何设备实时协作。

测试发现：The omnigent CLI reports version 0.13.0.dev0, the server listens on TCP and returns HTTP 200 on root/health/OpenAPI endpoints, and 929+ unit tests across 10 module directories pass without failure. 28 分钟。

可私有部署的多租户知识智能体平台：统一 RAG、知识图谱、多智能体、MCP/Skills、沙盒与权限管理。Yuxi = Cloud Agents + Knowledge RAG，一个自托管的知识 agent 平台，用于 RAG、知识图谱和多 agent 工作流。

测试发现：Backend Python dependencies installed and locked; all 1640 backend unit tests pass; all 164 web unit tests pass; Vite production build completes successfully; engineering trust contracts and lint checks pass. Two minor test permission-assertion bugs fixed (umask compatibility) and one Node.js version gap resolved... 25 分钟。

RAGFlow 是一个 agent 原生数据库：一个检索增强生成（RAG）引擎，已发展成上下文基础设施。

测试发现：Python 3.13 venv with 697 deps installed, Go toolchain auto-resolved to go1.27, C++ librag_tokenizer_c_api.a built with clang++-20, Go binaries ragflow_server and ragflow-cli compiled successfully, Go unit tests pass (all packages), Python unit tests pass (5145/5145), frontend builds successfully, native static libs... 40 分钟。

在你的机器上运行的求职搜索。基于 Claude Code 构建的 AI 求职申请框架：评估职位信息、定制简历、撰写求职信、准备面试。Fork 它并拥有它。

测试发现：Python test suite passes (805/805), all 6 portal CLI test suites pass (495/495), CV and cover letter compile to verified PDFs with lualatex/xelatex, and live job searches return real data from Danish and global job portals. 14 分钟。

在 dsh 里装上这个插件即可，无需登录、注册或填 API Key，就能使用包括 DeepSeek V4.1 Flash、Kimi K3 在内的前沿模型, 完全免费，不限量。 All you do is install this plugin in dsh: no login, no sign-up, no API...

测试发现：Node 22.23.3 installed via n; npm install added 21 dev packages; tsc --noEmit passed; all 31 test suites pass; host-selftest confirmed the plugin activates, serves model picker with 26 free models + 1 region-limited, performs plain chat, tool calling, effort budgets, vision, and OpenAI-compatible forwarding against... 16 分钟。

500 AI Agents Projects 是一个精选的 AI agent 用例集合，涵盖多个行业。它展示了实际应用，并提供开源项目链接用于实施，说明 AI agent 如何改变医疗、金融、教育、零售等行业。

测试发现：Web app builds with vite and serves HTTP 200 on port 5173. All 3 CrewAI MCP course lessons execute end-to-end with OpenRouter. SQL query agent creates demo database and answers natural language questions. Unit test generator agent produces passing 21/21 pytest suites. News summarizer falls back to mock data and... 12 分钟。

一个在容器中运行以确保安全的轻量级 OpenClaw 替代方案。连接 WhatsApp，Telegram，Slack，Discord，Gmail 和其他消息应用，，具有记忆，定时任务，并直接运行在 Anthropic 的 Agents SDK 上。

测试发现：The NanoClaw project installs, compiles with tsc, passes typecheck with tsc --noEmit, and all 2945 host tests pass via vitest. 10 分钟。

JavaScript 页面内 GUI agent。使用自然语言控制 Web 界面。

测试发现：Install, sequential build, typecheck, lint, and all 69 unit tests pass. The project is fully functional with Node v22. via nvm. 27 分钟。

Distilly：将他们的思考方式提炼为适用于任何 Agent 或 Bot 的可复用 Skills。原名为 Colleague Skill（原同事 Skill）。

测试发现：The full test suite passes (73/73, plus 9 subtests), the Node.js CLI installs the distilly skill to any target path, and every Python tool in tools/ prints correct usage on --help. 2 分钟。

一个包含 100+ 个专用 Claude Code subagent 的集合，涵盖广泛的开发用例。

测试发现：The repository clone is complete: 161 agent definitions with valid YAML frontmatter across 10 categories, a syntax-valid interactive installer (needs TTY), a fully operational subagent-catalog CLI skill with 12-hour TTL cache and fetch/search/list/invalidate commands, and all documentation (README, CONTRIBUTING... 5 分钟。

Dolt：用于数据的 Git

测试发现：Built dolt 2.3.5 from source: 'dolt version' prints 2.3.5, CLI sql/log/diff work, sql-server answered MySQL clients on TCP (create db, create tables, query, commit, branch, merge, history), and push/clone/pull over a file:// remote completed, with only the gpg-dependent unit test failing due to the absent system gpg... 37 分钟。

一个用于构建、编排和部署 AI agent 和多 agent 工作流的框架，支持 Python 和.NET。

测试发现：Python Agent Framework installs and its entire test suite passes across core, tools, providers (OpenAI, Anthropic, Gemini, Ollama, Mistral, Bedrock, Foundry), orchestrations, declarative, A2A, hosting, Redis, ChatKit, GitHub Copilot, and AG-UI packages. 28 分钟。

全栈开发者的 67 项专业技能。将 Claude Code 转变为你的专家级结对编程搭档。

测试发现：python3 scripts/validate-skills.py passes with 0 errors, make test runs 12/12 passing, and the site builds 99 static pages successfully via Node 20. 6 分钟。

基于开源 TypeScript AI Agent 框架构建的 AI Agent 工程平台

测试发现：Monorepo builds (29/29 packages), lints (0 errors, 5 warnings), and all 25 packages with runnable unit tests pass (~2776 tests). The only test failure is @voltagent/e2e which requires an unreachable PostgreSQL database. Two test files with a pre-existing ESM dependency conflict are excluded. 26 分钟。

HarnessRouter Community Edition：自托管、Apache-2.0 版的 agent 工具统一接口。通过一个 API 运行 Codex、Claude Code、Hermes、PI、DSH 等，支持会话、流式传输、文件、取消和失败处理。实现了开放标准 Unified Harness...

测试发现：Gateway and runner servers start and respond to health checks; all three test suites (gateway: 622/17, runner: 490/2, protocol conformance: 85/2) pass. 6 分钟。

2,884+ 个可复用的 AI agent 技能，适用于 Claude Code、Codex、Cursor 和 Gemini CLI。AAS Core 增加了本地目录发现、agent 自主选择和计划预览。支持 CLI、MCP 和插件。

测试发现：With Node 22.23.4 and PyYAML installed into /tmp, dependency install, the 2459-skill validation, the 241 web-app tests, the web-app build and the catalog build all succeed, the catalog app serves HTTP 200 for its main routes, the AAS CLI creates and validates a stack manifest and its MCP server answers initialize and... 21 分钟。

供 agent 使用的脚本，在我的仓库之间共享。

测试发现：All 12 test suites pass; npm install completes with 0 vulnerabilities; Python, Node.js, and Ruby scripts compile and run correctly; browser-tools CLI loads and reports help; docs-list enumerates all docs with metadata checks; no Chrome binary available for browser launch tests. 26 分钟。

面向生产环境 AI 的多 agent harness

测试发现：With the fixes, ./quickstart.sh completes and launches the Hive dashboard on an available local port; /api/health returned 200, Chromium rendered the Hive landing UI without page errors, and make test exited 0 with 2496 core and 4381 tools tests passed. 21 分钟。

编排一切，无处不在。用于跨数据、基础设施、业务流程和 AI agents 的工作流的开源控制平面。声明式 YAML、事件驱动、支持任何语言、Git 版本化，可扩展到数百万次执行。

测试发现：Kestra v2.1.0-SNAPSHOT builds and tests successfully: backend core (3636/3646 pass, 2 environment-limited failures), executor (339/339), scheduler, worker (107/107), jdbc-h2 runner (98/98), and frontend (2743 unit tests + 282 storybook tests + type check + lint) all pass. 75 分钟。

智能技能路由与工作流编排，面向 AI agent：在 SkillsBench 上使用 DeepSeekV4Flash-VE 时，+21.12 pp 奖励，tokens −29.6%。

测试发现：VibeSkills v4.1.0 installs successfully, check verifies all 292 receipt-owned files, the vgo_cli Python launcher works end-to-end, and 129+ core unit tests pass with 3 isolated test failures due to missing pwsh in this container. 37 分钟。

Gentle-AI 配置你已经在使用的 AI coding agents：Claude Code、Cursor、OpenCode、Codex、Pi 等。可选择持久记忆、Organic-Driven Development、精选技能、MCP 服务器、角色以及可选的有界审查。开源，无 agent 锁定。

测试发现：gentle-ai 2.0.0 builds from source, prints version (2.0.0-20260906031619-2c25e878eea4) and help text, all 53 Go test packages pass. 17 分钟。

面向人文社会科学的 multi-agent 研究系统。

测试发现：MISAKA 0.18.11 installs and launches: --version prints 0.18.11, init creates a git project, the panel opens the setup wizard with fd and pdftotext present from the bundled archive, dm runs real turns that reply, and the test suite passes 4361 tests (the single remaining failure needs the optional ocrmypdf system... 22 分钟。

用于 Claude Code 的自主 AI 开发循环，具备智能退出检测

测试发现：Ralph is installed globally with 7 CLI commands, all 1319 tests pass, project setup works, and dry-run mode simulates loops without API calls. 13 分钟。

用于 Web 自动化的开源 AI agent。使用你真实且已登录的浏览器来自动化日常任务。

测试发现：Dependencies installed, project builds to dist/ (background.iife.js 1,848.81 kB, manifest.json), TypeScript type-check passes across all 15 packages, unit tests pass (14/14). 15 分钟。

AutoClip｜一个链接，一键出片。开源 AI 视频剪辑桌面工具，将播客、访谈、课程等长视频自动剪成短视频，生成字幕、封面和发布文案，适配抖音、小红书、TikTok、Reels 与 YouTube Shorts。开源 AI 视频剪辑与内容再利用。

测试发现：The AutoClip FastAPI backend serves all API health endpoints and the test suite passes with 103/103 tests passing. The frontend builds to production with Vite. 4 分钟。

PageIndex：面向无向量、基于推理的 RAG 的文档索引

测试发现：PageIndex SDK v0.2.10 installed in a venv, all 796 unit tests pass, the CLI parses correctly, and the package imports without error. 3 分钟。

277 个即插即用的 AI 专家角色，支持 Claude Code/Cursor/Copilot 等 20 种工具，覆盖工程/设计/营销/金融等 20 个部门。含 64 个中国市场原创智能体（小红书/抖音/微信/飞书/钉钉/Qt 上位机/机械设计）。搭配编排器...

测试发现：All 277 AI agent definition files are valid, the npm package builds, the conversion pipeline produces output for all 18 supported tools, and the install script copies agents to 6 detected tool directories without errors. 5 分钟。

Hermes WebUI：通过 Web 或手机使用 Hermes Agent 的最佳方式。

测试发现：Hermes WebUI boots from .venv, serves /health (HTTP 200), /api/onboarding/status returns full JSON, and 16893 of 17245 pytest tests pass. The missing Hermes agent (not installed) is the only missing component for full functionality. 38 分钟。

AG-UI：Agent-用户交互协议。将 Agents 引入前端应用。

测试发现：All 35 Nx workspace projects build and 25/26 test suites pass (5531 tests), the dojo Next.js app serves HTTP 200 on localhost:3000. 14 分钟。

开源、自托管的 Claude Code：一个终端 AI 助手，以及其背后的 Python 框架。工具调用、沙盒执行、多 agent 团队、技能、检查点、无限上下文，基于 Pydantic AI，任何模型。

测试发现：Package imports cleanly, CLI runs and reports version v0.3.43, 2771 of 2785 tests pass with pydantic-ai-backend pinned to v0.2.29. 25 分钟。

一个用于 agent 、工具和服务的框架

测试发现：Go Micro builds and tests run with the CLI binary working; README updated to satisfy first-agent wayfinding doc integrity tests that enforce install-troubleshooting, micro agent demo/quickcheck/examples/zero-to-hero, examples/INDEX.md, first-agent, support, no-secret, your-first-agent, debugging-agents, inspect, and... 31 分钟。

agent harness 性能优化系统，为 Claude Code、Codex、Opencode、Cursor 等提供技能、直觉、记忆、安全和研究优先的开发。

测试发现：npm install completed, all 10 CI validators pass, 4876 of 4906 tests pass; the only fix applied was guarding plan.adapter access in codex-user-config.js and ownership-guard.js to prevent TypeError when tests call applyInstallPlan without an adapter field. 42 分钟。

Python 如何做 AI。agents、实时语音、图像生成、嵌入。每种模型、每种接口，全程类型安全。

测试发现：The workspace installs and builds, the full CI-configured suite passed 15182 tests, and clai web served HTTP 200 on its health, configure, index, and chat endpoints using the built-in test model. 19 分钟。

面向大厂Agent研发岗位求职的agent教程网站，涵盖技术路线与面试八股文

测试发现：Agent API Lab Python test suite passes 18/18. All 10 run_lab.py scenarios produce correct deterministic output with expected statuses (completed, stuck, protocol_error, refused, truncated). Static HTTP server serves index.html (200, 21KB), CSS, JS, assets, and all 168 article markdown files. git diff --check clean... 7 分钟。

将工程纪律安装到任何 AI 编码助手。可组合的技能涵盖设计、实现、审查和团队标准。更好的流程，而不仅仅是更好的提示。

测试发现：The Lattice skils install, distribution rebuild, and the portable verification runner all run correctly (27 skills installed, runner passes and fails as specified), while the website build remains blocked on the container's Node 18 versus the repo's pinned Node >=20 requirement. 15 分钟。

将任何代码库（包括其文档、SQL 模式、配置和 PDF）转化为可查询的知识图谱。适用于 Claude Code、Cursor、Codex 和 Gemini CLI 的 /graphify 技能：本地确定性 AST 解析，每条边都有解释，没有向量存储。

测试发现：graphifyy CLI installs, builds knowledge graphs from source code locally with tree-sitter AST, and passes 5234/5288 tests (11 pre-existing skillgen failures from shallow clone, no fix available). 11 分钟。

开源 Python 库，用于 AI 浏览器 agents，以及 Browser Use Cloud：隐身浏览器和托管 agent API。

测试发现：browser-use v0.13.10 installed in a Python 3.12 venv with uv, Chromium headless installed via playwright, and all 563 CI tests pass (529 passed, 34 skipped) after fixing 2 test identity comparison bugs in test_beta_agent.py that only manifest under xdist worker parallelism. 75 分钟。

将任何网站变成 CLI，并通过 AI agent 使用你已登录的浏览器。

测试发现：OpenCLI 1.8.8 builds and runs on Node 22.14.0; 632 test files pass (7325 tests, 3 skipped) across unit, extension, adapter, smoke projects; CLI binary answers version 1.8.8 and lists 1366 registered commands. 27 分钟。

用于编排角色扮演、自主 AI agent 的框架。通过培养协作智能，CrewAI 使 agent 能够无缝协作，处理复杂任务。

测试发现：CrewAI v1.15.22 is installed with all optional extras, the CLI reports version 1.15.22, and 5424 of 5427 crewai tests pass (3 pre-existing unrelated failures), 87 crewai-core tests pass, 489 crewai-tools tests pass. 59 分钟。

使用 AI 设计实体产品

测试发现：cadgen 0.6.6 installed from source: OCP kernel loads, CLI responds, Python test suites for cadgen CLI/store/memo and all eight tested skill packages pass, cross-language JS runtime bundle and Playwright snapshot browser present. 32 分钟。

一个开源的、代码优先的 Python 工具包，用于构建、评估和部署复杂的 AI agents，兼具灵活性和控制力。

测试发现：uv virtual environment installed with all extras, google-adk 2.9.2 builds and imports, full unit test suite passes at 16167/16167 (0 failures, 83 skipped, 26 expected failures, 2 unexpected passes), CLICK-based CLI entry point is functional. 14 分钟。

三省六部制 · OpenClaw Multi-Agent Orchestration System：9 个专业化的 AI agent，具备实时仪表板、模型配置和完整的审计轨迹

测试发现：All 68 tests pass on Python 3.12.3; the dashboard HTTP server serves the React frontend, health check, and all API endpoints correctly. 6 分钟。

MiMo Code：模型与 Agents 共同进化

测试发现：MiMoCode CLI runs, typechecks, builds SDK, and passes ~2000+ tests across all core packages (agent, config, auth, git, tool, session, cli, server, workflow, effect, actor, history) with only ~15 environmentally-specific failures (umask, Bun version, test isolation). 37 分钟。

编码、构建和评估 agents：出色的模型与技能/MCP/ACP/A2A 支持。

测试发现：All 8730 unit tests pass, CLI boots and responds, playback LLM agent runs end-to-end, format/lint/typecheck all pass. 30 分钟。

桌面 AI agent 会为每项任务雇用专家，能在您的真实应用中持续工作数周，并接受来自手机的指令。免费，Apache-2.0。

测试发现：Agentlas OS runtime v1.2.57 installed to ~/.agentlas/runtime/current with hephaestus doctor ok; ontology runtime operational with SQLite storage, Model2Vec int8 embeddings, FTS adapters, and auto-ingest working from ~/.agentlas/inbox directories. 6 分钟。

TypeScript AI SDK 和 agent 框架，零依赖。构建生产级 AI agent 和多 agent 集群：持久执行、长期记忆、混合 RAG、MCP 客户端/服务器、A2A、人工审批、CodeAct 沙箱、实时语音。一个流式 API，29...

测试发现：Both @deuz-sdk/core and @deuz-sdk/react build from source and pass their full test suites (2635+39 tests passing) using MSW-mocked HTTP providers, requiring no real API credentials. 7 分钟。

一个透明、极简、可定制的 agent 框架。约 300 行可读代码。完全控制，无魔法。

测试发现：All 30 tests pass, core Agent/Runner/Session/tool imports work, and the evaluation module loads successfully. 2 分钟。

梯度下降，但参数是 agent：一个用于自我进化 agent（技能、提示词、harness）的并行异步框架。差异就是梯度；聚合器就是优化器。

测试发现：AgentDescent 0.5.0 is installed in a Python 3.12 venv; all 3026 tests pass; the demo and CLI (demo, init, plan, evolve, etc.) run offline with no API key needed. 48 分钟。

构建一个 agent harness 并对其进行端到端控制。用于生产环境 AI agent 的开源 SDK，支持 Python 和 TypeScript，兼容任意模型、任意云。

测试发现：Python SDK strands-agents v0.1.dev1 and strands-harness v0.0.1.dev1 installed from source, both packages import correctly, and over 2,000 unit tests pass across both packages. 68 分钟。

Cognee 是面向 agent 的开源 AI 记忆平台。使用小模型即可免费为你的 AI agent 提供持久长期记忆。

测试发现：cognee 1.6.1 installs, builds, starts, and runs: the local LLM-free remember/recall pipeline produces search results from ingested text, the CLI demo loads a bundled knowledge graph, the FastAPI server answers HTTP 200 on /health, and 388 unit tests pass across 5 test suites with zero failures. 33 分钟。

用于语义搜索、LLM 编排和语言模型工作流的一体化 AI 框架

测试发现：txtai 9.14.0 installed in venv, all core embeddings/graph/workflow/agent/cloud/vector/scoring tests pass, API responds to search/index/count endpoints with real model inference, ~470 automated tests pass across 30 test suites. 83 分钟。

官方 Crawl4AI Cloud：一个面向 AI agents 的托管网页抓取、提取和搜索 API。免费 API 密钥，每月 1,000 积分，无需信用卡。完全开源：这个 repo 就是驱动它的 Apache-2.0 爬虫，你可以自托管它。

测试发现：Crawl4AI 0.9.4 installed in venv, Playwright chromium browser downloaded, DB initialized, AsyncWebCrawler can crawl real URLs with markdown output, cache operations work, and 360 core tests pass with 11 skipped. 14 分钟。

使用 Python 构建自主 AI agent。

测试发现：Upsonic v0.77.3 installed from source in editable mode; all 1490+ tests pass; CLI is functional. 23 分钟。

使用 agents 对任何内容进行逆向工程，从应用行为到原生二进制文件。

测试发现：npm ci installs 281 packages, tsc builds cleanly to dist/, and the full vitest suite runs 353 test files (1756 tests) with zero failures on Node 22.19.0. 7 分钟。

The Memory Layer for AI Agents：为 AI 代理和应用提供即插即用的记忆基础设施，具备持久化的上下文，专为生产环境构建。

测试发现：mem0ai 2.2.0 installed and tested: 1210 tests passing across the core SDK, LLM providers, embeddings, client, telemetry, proxy, configs, and Qdrant vector store, with one bugfix in proxy/main.py to guard against UnboundLocalError on memories_text. 23 分钟。

面向 Claude Code 和 AI agent 的营销技能。涵盖 CRO、文案写作、SEO、分析和增长工程。

测试发现：All 50 skills validated against the Agent Skills spec (both internal and official skills-ref validators), all 26 CLI contract-test suites pass with mocked HTTP, all 65 CLI tools parse cleanly, all 3 JSON manifests are well-formed, and skillkit can enumerate 53 available skills from the repo. 5 分钟。

DeepTutor：终身个性化辅导。

测试发现：Package installed via pip install -e .[dev] with partners extra. LLM features require OpenAI/etc credentials (deeptutor doctor reports No active LLM model). Test suite passes across core, API, CLI, services, book, reading, video learning, agents, capabilities, tools, knowledge, runtime, and plugins modules after 4... 81 分钟。

你的轻量级开源个人 agent。

测试发现：After pinning mcp<1.30 and relaxing two width-sensitive CLI assertions, Python installs editable, the 6785-test suite, 1125 WebUI tests and basedpyright all pass, and against a local mock provider the CLI agent, OpenAI-compatible API server (JSON and SSE), gateway WebSocket chat and native TUI all returned 'Hello from... 29 分钟。

将 Claude Code 变成完整的游戏开发工作室：49 个 AI agent，72 个工作流技能，以及一个完整的协调系统，模拟真实工作室的层级结构。

测试发现：The Claude Code Game Studios framework repository is fully verified: all 14 hooks run cleanly (session-start reports review mode solo and detects fresh project, detect-gaps identifies new project and suggests /start, validate-* hooks exit 0 on non-matching context, yaml-helper resolves config correctly), all 8 scripts... 4 分钟。

ARTEMIS 将自然语言指令转化为可靠的 Android 自动化。它自动化端到端工作流，捕获 logs，并与 Antigravity、Codex 和 Claude Code 等 AI 编码助手无缝集成。它在 AndroidWorld Benchmark 上达到 99%+ 的成功率。

测试发现：uv sync --dev succeeded; pytest exits 0 with 2185 passed, 10 skipped, 8 deselected; 'artemis ui' served the web console with HTTP 200 on /, /admin, /docs, and /health; --version answered v1.0; doctor and status CLI commands run, and the ADB daemon started from the bundled binary. 61 分钟。

开源、安全的环境，配备面向企业级 agent 的真实工具。

测试发现：The repository installs and builds with the pinned toolchain, the CLI launches at version 2.20.0, the no-credential CLI and Python offline suites pass, and an isolated installed CLI completed a mocked template build including a COPY upload. 10 分钟。

Agent S：最先进的开源计算机使用 agents

测试发现：gui-agents v0.3.2 installed and its 5 unit tests pass; all S1/S2/S2.5/S3 modules import and engine constructors work correctly. 8 分钟。

面向 LLM 与 AI Agents 的自进化记忆操作系统：超持久记忆、混合检索与跨任务技能复用，可节省 35.24% 的 token，并支持 DeepSeek Harness。

测试发现：Python 3.12.3 venv with Poetry, MemoryOS 2.0.33 installed, library imports, CLI runs, all 875 tests pass with 9 skipped for expected missing infrastructure (Neo4j, Redis). 14 分钟。

一体化 LLM CLI 工具，具有 Shell Assistant、Chat-REPL、RAG、AI 工具与 Agents，支持 OpenAI、Claude、Gemini、Ollama、Groq 等。

测试发现：aichat 0.30.0 builds from source, passes all 21 unit tests, and successfully proxies requests to an OpenAI-compatible API backend including chat completions (200), model listing, and web playground (200). 7 分钟。

私有 AI agents，适用于高度受监管的团队，也适用于任何希望在自己的电脑上使用 AI 的人。开源 Python agent 框架，可完全离线运行。

测试发现：Both praisonaiagents and PraisonAI packages install and import successfully. Agent creation works. All core agent unit tests pass (480/480). LLM token tracking race condition fixed with ContextVar. LLM class supports __deepcopy__ for agent cloning. Knowledge tests pass with chonkie installed. 193/194 LLM tests pass... 32 分钟。

Android 全栈设备控制平台：WebRTC/H.264 远程桌面、UI/OCR/图像匹配自动化、一键 MITM、内置 Frida、代理/VPN/frp/P2P 网络、MCP/Agent、160+ 个 API，专为多设备集群和工程化部署而设计。

测试发现：lamda Python client v10.8 installs, all modules import, and the gRPC protocol stack works end-to-end via a mocked server. 7 分钟。

使用 Model Context Protocol 和简单工作流模式构建高效的 agent。

测试发现：mcp-agent 0.2.6 installs, builds, CLI boots, and all 1501 tests pass after fixing the abstract generate_stream method in the base AugmentedLLM class and patching the Bedrock streaming test mock. 20 分钟。

一个三语（繁中 / English / 简中）的 agentic AI 学习路线图：从 LLM 基础到 multi-agent 系统，包含 240+ 精选资源和动手示例。中文 AI agent 学习地图。

测试发现：The mkdocs site builds to _build/site/, the mdBook builds to book/dist/, and all 25 example test suites (stages 1-7) pass with mock/offline mocks, no real API keys required. 21 分钟。

MiroThinker 是一个深度研究 agent，针对复杂研究和预测任务进行了优化。我们最新的模型 MiroThinker-1.7 在 BrowseComp 和 BrowseComp Zh 上分别取得了 74.0 和 75.3。

测试发现：All 5 packages install and all source modules import correctly. The agent pipeline executes end-to-end with a mock LLM server: connects to MCP tool servers, calls LLM, processes response, writes task log with status=success. Flask trace visualizer answers HTTP 200. 12 分钟。

AI 原生网络安全的行动系统：意图转化为受治理的执行，证据转化为运营记忆，每一次操作都会改进下一次。

测试发现：The CyberStrikeAI binary builds cleanly with Go 1.25.0, serves its web UI on HTTP port 8080 with a 200 status, accepts admin login via POST /api/auth/login returning a full permissions token, and all 28 internal Go test packages (including database, handler, multiagent, security, mcp, and workflow) pass without... 17 分钟。

自托管的 TypeScript agent 运行时，具备持久审批和可验证的运行记录。拥有它，批准它，审计它。

测试发现：The monorepo installs, builds, passes 2350 unit tests across all 4 workspaces plus the benchmark harness, and the CLI prints help, all with Node.js 22, no API keys or network required. 4 分钟。

IntentKit 是一个开源、自托管的云 agent 集群，为你管理一个协作式 AI agent 团队。

测试发现：Project installs with uv, the intentkit Python package and FastAPI server import correctly, and 1760 of 1774 non-BDD/non-migration tests pass using SQLite in-memory engine as a PostgreSQL replacement inside the container. 56 分钟。

5+ 个 benchmark 上的 Top-1 | Web UI | 支持 MiroThinker、Claude、Kimi、OpenAI

测试发现：MiroFlow agent pipeline runs end-to-end with a mock OpenRouter API: uv sync installs all deps, the agent loads config, initializes the LLM client, calls the API, processes tool-using context, and extracts a boxed final answer. 10 分钟。

全栈 AI 应用生成器：FastAPI + Next.js，内置 AI Agents、RAG、流式输出、身份验证和 20 多种集成，开箱即用。

测试发现：All 681 tests pass; the CLI generates projects and runs type-checking on all generated matrix configurations including pydantic_deep_pg, pydantic_ai_code_execution, and pydantic_ai_skills without errors. 37 分钟。

用于下一代视频交互和工作流的 AI video agents 框架。

测试发现：Backend serves all 25 agent listings, config check, session CRUD, and collection endpoints via a mock VideoDB server. Frontend dev server builds and serves the Vue app. Both run simultaneously on :8000 and :8080. 16 分钟。

适用于 Azure 上 AI agent 应用的企业级着陆区。可部署安全基础设施、agent 应用或两者兼有：基于 Microsoft Foundry 和 Microsoft Agent Framework 构建，具备零信任网络、可观测性和 IaC。

测试发现：The repository installs into a clean venv and its full unit/contract suite passes (469 passed, 431 subtests) with real module CLIs resolving topology/composition and the thin hosted-prepare hook running to its documented classic result, while Azure-only paths fail closed with the documented exit codes. 5 分钟。

适用于 Azure 上 AI agent 应用程序的企业级登陆区域。部署安全基础设施、agent 应用，或两者；基于 Microsoft Foundry 和 Microsoft Agent Framework 构建，具备零信任网络、可观测性和基础设施即代码。

测试发现：In a fresh container venv, the full pytest suite passes (469 passed, 9 environment skips), all five bash hooks pass syntax checks, and the topology, hosted-access, and app-config CLIs run to completion against mocked Azure and fail closed without it. 6 分钟。

不是问答机器人，而是能真正干活的数字分身。共享智能体，共享项目上下文，让项目所有成员一起协同工作。同时智能体会像人一样通过资料或者工作抽象和总结经验到项目上下文中

测试发现：Backend API runs on port 8002 (SQLite, auth-bypass) with /health 200, project+knowledge APIs working, and a real agent turn streamed an SSE-mocked response end-to-end; 7 agent-core sandbox/shell tests remain red in this container due to missing root for python symlink/user-site and venv path layout. 37 分钟。

Blades 是一个基于 Go 的多模态 AI Agent 框架。

测试发现：Go installed user-space, all modules build, all 29 test packages pass, and the blades CLI init/doctor/run commands work with blades run answering correctly against a local OpenAI-compatible mock. 17 分钟。

又一个 coding agent harness，轻量级，用 go 编写。

测试发现：zot 0.0.0 builds from source under a toolchain I installed, prints its version, launches its TUI, and ran a complete streaming agent conversation against my local OpenAI-compatible mock with exit 0, writing a session file; the full test suite passes 16/16 under -race when the container's injected OPENROUTER_API_KEY is... 22 分钟。

👾 下一代透明智能体架构 | Next-Gen Transparent Agent Architecture 🔍 全行为审计 | 🛡️ 两段式安全调用 | 🧠 双水位记忆 | ⏰ 心跳任务 📊 P0 级事故率降低 80% | 兼容 OpenClaw + Claude Code 技能生态

测试发现：Installed and the 130 tests all pass. The CLI starts, connects to an LLM provider, completes a tool-augmented conversation (user query -> get_current_time tool call -> mock response displayed), and exits cleanly. 4 分钟。

以 Laravel 方式构建、测试和部署智能 AI agents。

测试发现：PHP 8.4.26 and Composer 2.10.3 installed via static binaries, all 136 Composer dependencies resolved and autoloaded, and the Pest test suite passes fully (1084/1084 passing tests) using SQLite in-memory database and mock LLM providers. 9 分钟。

开源实时语音 agent 服务器，用 Go 编写，支持 WebRTC (WHIP)、barge-in、流式 STT/LLM/TTS 管道、插件系统、多语言 SDK、SIP 电话、ESP32 支持以及完全本地模式。

测试发现：The Go server builds, lints with no formatting issues, passes all 15 test suites, and starts on :8080 responding to /health and /whip endpoints. 5 分钟。

一个开源的 Digital Worker 平台，用于可靠执行、持续协同演进和构建企业 AI 资产。

测试发现：Core Python service tests pass (1564/1564), Go backend builds and passes 44 test packages, frontend SPA builds and app tests pass (161/161). 44 分钟。

SmythOS Runtime Environment (SRE) 是一个开源的云原生运行时，用于 agentic AI。它安全、模块化且可用于生产环境，让开发者能够在本地、云端和边缘环境中构建、运行和管理智能 agent。

测试发现：SmythOS SRE v0.5.0 monorepo installs with pnpm; CLI reports version 0.4.0 and responds to commands; core unit test suite passes 368/377 tests (19 failures require real AWS/OpenAI API credentials which cannot be mocked); SRE runtime initializes with Storage (LocalStorage) and Cache (RAM) connectors operational... 27 分钟。

Parlant 是一个用于构建面向客户的 AI 聊天 agent 的开放平台。描述你希望 agent 如何表现，你的 coding agent 就会为你实现它。

测试发现：With a schema-correct mock NLP backend, the Parlant server started on port 18999, answered /healthz 200, created agents/guidelines/sessions (201), and ran the engine to completion status; 208 API tests passed while the parlant CLI remained unimportable. 54 分钟。

Harness 工程初学者教程，从 0 到 1

测试发现：Root npm install succeeded, VitePress docs dev server serves all pages with 200, all 15 project Electron apps build successfully, and project-01 starter Electron app launches and stays running in the container. 15 分钟。

自动驾驶 agent 平台

测试发现：Agno SDK v3.0.11 installed and imported; CLI functional; core agent/context/reasoning/utils/model/tool test suites (5000+ tests) pass; format script passes; validate passes ruff checks but has 60 pre-existing mypy errors in the agno library. 62 分钟。

使用 AI agents 设计和审查软件的开源技能

测试发现：All 49 unit tests pass across 6 test modules: frontmatter validation, link integrity, harness/plugin manifests, scenario coverage, skill load limits, and Claude trace directory parsing. 1 分钟。

一个自我进化的 agentic OS。一种基于操作系统原理构建的 agent 执行新架构，将 harness 作为通向超级智能的路径。

测试发现：All Linux-compatible source checks, 76/76 Rust contract tests, and 60+ client contract tests pass; CLI binary answers help and status; macOS-native features (runtime daemon, CLI build, native filesystem module) are unavailable on Linux as designed. 27 分钟。

轻松搜索、理解、复现并改进一个想法

测试发现：Both aworld and aworld-cli packages install and build cleanly; the aworld API server starts and serves /docs (HTTP 200); the web server starts (HTTP 307); the CLI help prints correctly; 413+ unit tests pass covering ACP protocol, CLI memory, config, utils, and context modules; one import bug (missing logger) was fixed... 23 分钟。

一个用于构建可组合 AI Agent 的生产就绪框架

测试发现：Alphora 1.3.4 is installed in a venv, all 22 tutorials run without errors, the alphora-web frontend server responds to HTTP 200 on /v1/health, and the publish_agent_api server accepts streaming and non-streaming chat completion requests against a mock OpenAI backend. 17 分钟。

从 Claude Code、Codex 或任何 coding agent 中获得 10X 的收益

测试发现：Rust workspace (excluding tauri-app) compiles and passes cargo check. Server binary builds and starts, binding to random available port and responding to HTTP requests. 3 crate test suites pass (10/10 unit tests). Frontend TypeScript type checks pass for web-core and local-web. The npx CLI, Tauri desktop app, and web... 82 分钟。

最具 Token 效率的开源 AI Agent

测试发现：无法验证；log 显示了它停在哪里。 87 分钟。

开源 AI 同事各自拥有一台专属计算机：浏览器、文件和工具，所有操作都在执行前决定，并在执行后记录。可接入任意 AG-UI agent。

测试发现：bun install and all 3985 non-integration unit tests pass. The 69 failures are all infrastructure-gated (no PostgreSQL for integration tests, no Docker for compose tests, 2 pre-existing Mastra .env leakage failures). 49 分钟。

计划、运行并监督从任务到合并 PR 的 agent 集群。桌面端（macOS、Windows、Linux）和 iPhone 与 Android 应用。支持 Claude Code、Codex 和 35+ 个 coding agent。

测试发现：无法验证；log 显示了它停在哪里。

Astrid 是一个可移植的、能力安全的操作系统，用于可组合软件。

测试发现：无法验证；log 显示了它停在哪里。 72 分钟。

AgenticX 是一个统一的生产级多 agent 平台，包含 Python SDK + CLI (agx) + Studio server + Machi 桌面应用。功能包括 Meta-Agent 编排、15 个以上 LLM 提供商、MCP Hub、分层记忆、头像与群聊、技能生态系统、安全沙箱以及 IM...

测试发现：无法验证；log 显示了它停在哪里。

自托管、开源的 agent 技能注册表，适用于企业。发布并版本化管理技能软件包，通过 RBAC 和审计日志进行治理，可使用 Docker 或 Kubernetes 进行本地部署。

测试发现：All 4 backend Maven modules compile and 1483 of 1489 tests pass (6 fail only due to Docker unavailability). Frontend dependencies install and individual vitest tests pass. The full app cannot start because Docker is not available to launch Postgres, Redis, and MinIO. 88 分钟。