在全新机器上运行过的开源 coding agents

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。

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。

一个用于结构化接口的 JavaScript 库，具有显式的所有权和生命周期。专为 agent 主导的开发而构建。Canonical Marionette 仓库；v4 历史保留在 v4 分支上。

测试发现：Marionette v5.0.0 builds and all 2114 unit tests pass on Node 24.19.0 with npm 11.17.0. 4 分钟。

一个免费的 coding agent，运行在你的终端中。它运行在当下最智能的免费模型上。

测试发现：The ANUS CLI installs cleanly, all 28 node --test tests pass, and the --version flag returns 0.2.2. 2 分钟。

一个能了解你的 agent 在做什么的终端窗口管理器。支持平铺窗格、工作区、重启后依然保留的会话，以及每个 coding agent 一个 Inbox。官方渠道仅限 tuios.dev 和本 repo。tuios 没有 X/Twitter 账号。

测试发现：tuios and tuios-web binaries build from source, the full Go test suite (49 packages) passes, and the daemon creates sessions, accepts typed input via send-text/send-keys, and captures pane output showing shell command execution. 22 分钟。

一个开源的氛围编程平台，帮助你构建自己的氛围编程平台，完全构建在 Cloudflare 技术栈上。

测试发现：The VibeSDK project installs, builds, typechecks, and passes all 514 unit tests (root vitest + SDK bun tests) using Bun 1.4.2 and Node.js 20.18.0 inside the container. 11 分钟。

面向 Coding Agents 的经验驱动黑盒成本控制

测试发现：tokeneconomizer package installed in venv, all 420 tests pass, te CLI outputs status and doctor diagnostics, and the plugin bundle builds to /tmp/test_bundle. 6 分钟。

开源 AI 视频 agent：将任何 coding agent 变成视频 agent，并带有编辑器，可更改它生成的任何内容。

测试发现：Node 22.11.0 installed. pnpm 9.15.9 works. Playwright Chromium 149.0.7827.55 + headless shell installed. All 869 of 871 tests pass (2 skipped). CLI launches (version 0.1.6). Film creation/look/validation work end-to-end: real Chromium frames drawn and verified pure by the look command. 21 分钟。

面向 coding agents 的本地视频编辑器：42 个 FFmpeg 工具，带有机器可读的合约和一个 MCP 服务器，适用于 Claude Code、Cursor 和 Codex。剪辑、字幕、重新构图、音量、平台导出。无需云服务，无需 API 密钥。

测试发现：ffmpeg-skill 2.10.3 installed cleanly into /tmp/skilltest/ffmpeg-skill, doctor reports 67 capabilities with 0 required missing, the 169-test contract suite exits OK, and a synthetic clip was cut, probed at 75 frames/2.52s and decoded with no ffmpeg errors. 25 分钟。

Antigravity, Claude Code, Codex, Copilot, Gemini, Muse, OpenCode, Pi, and Prime Agent versions of Poteto's pstack. Rigorous agent workflows with Cursor...

测试发现：pstack v0.9.45 installs cleanly: bun tools/generate.mjs passes all 16 validation checks, bun test passes 227/227 tests, markdownlint passes on 165 files with 0 errors, the skills CLI installs all 54 skills and diff matches the source tree byte-for-byte, the vendored scripts compile cleanly, and the working tree is... 3 分钟。

适用于 VS Code 的开源 AI 编码助手。Coding agent、代码补全、聊天、编辑和审查，支持本地或托管模型。你的模型，你的基础设施。

测试发现：npm install completed, TypeScript compilation and esbuild bundling succeed, the full VS Code extension test suite passes (642 passed, 0 failed), and both twinny-node and twinny-server CLI binaries respond correctly. 8 分钟。

开源托管 agent 平台。可自托管，是 Claude Managed Agents 和 OpenAI's Agents API 的替代方案。

测试发现：Omnara builds, all 124 unit test packages pass, all 34 integration test packages pass against live PostgreSQL 18 + Redis 7 + MinIO S3, all 4 server binaries build and the API binary starts and responds on port 8080, and the frontend web app builds successfully. 43 分钟。

用于跟踪你的 repos 中 AI 生成的代码的 Git 扩展

测试发现：git-ai builds from source with cargo, all 2481 unit tests pass, 1310+ sampled integration tests across checkpointing, blame, diff, rewrite ops, agent presets, worktrees, and async daemon mode all pass with RUSTUP_TOOLCHAIN=stable and GIT_AI_TEST_BINARY_PATH set. 45 分钟。

一个面向 AI coding agent 的模型驱动 harness：语义建模、应用生成、渐进式 API Assist 和基于证据的验证。

测试发现：Python-based teaql_workspace.py tool verifies and applies runtime source configurations for all 7 supported languages; the build-teaql-app skill, golden examples, and reference documentation are complete and well-formed; the npx skills add install path is gated by a Node.js 22 requirement not met in this container. 3 分钟。

面向 AI coding agents 的精选技能集合。技能是打包的指令和脚本，用于扩展 agent 在开发、文档、规划和专业工作流程中的能力。

测试发现：The build script assembles all 56 plugins into dist/ (232 files, 1.7MB), the marketplace.json passes CI validation, dist is in sync with source, version bumping works, and every SKILL.md, Python script, and bash script in the repository is syntactically valid. 4 分钟。

以 100 倍的速度审查代码。Lens 在拉取请求内部将每个 PR 绘制为动画架构和数据流导览。可将其用作 GitHub App、GitHub Action、CLI 或编码 agent 的 Skill。

测试发现：872 tests pass across all 7 workspace packages, build and typecheck are clean, CLI binary runs and prints version 0.11.0. 3 分钟。

Fable Workflow：将 Claude Fable 5 的工作方式提炼为任何模型都能运行的技能，并附有保持其真实性的评估。思考 / 行动 / 证明。

测试发现：The Fable Method repository installs cleanly, its 4 skills (fable-method, fable-loop, fable-judge, fable-domain) are at ~/.claude/skills/ with valid frontmatter, 9 domain adapters and 14 eval scenarios are complete, 15 eval result JSON files parse, and all Python/JS eval fixtures execute as designed. 3 分钟。

桌面应用和 Web 服务器，用于浏览、搜索和分析你的 AI 编码助手历史记录：Claude Code、Codex CLI、GitHub Copilot、Gemini CLI、Cursor、OpenCode 等。100% 离线。

测试发现：The WebUI server binary builds, starts, and serves the full frontend HTML on port 3727 with a health endpoint returning {'status':'ok'} and HTTP 200 on every route; all 1392 frontend JS tests and 1111 Rust library tests pass. 43 分钟。

适用于 Windows 和 macOS 的开源 AI agent 桌面应用，支持一键安装 Claude Code、MCP 工具和 Skills，并具备沙箱隔离、多模型支持以及 Feishu/Slack 集成。

测试发现：npm install completed successfully with Node.js v22.22.0. 155 of 158 test files pass (1141 of 1144 tests), MCP servers are built, better-sqlite3 native module rebuilt, and Electron v41.7.1 is ready. 63 分钟。

一个开源的 coding agent，运行在你的终端中，由 MiniMax 驱动。

测试发现：MiniMax Code 0.5.5 builds from source, passes 6003 of 6007 tests across smoke, BYOK, release-tools, status-contract, policy, sandbox, artifact, and capability suites, and answers to mcode --version. 27 分钟。

Nimbalyst - 面向 Claude Code、Codex 和 OpenCode 的开源可视化工作区。可并行运行多个 coding agent，在 Markdown、mockup 和图表中直观编辑其工作，并跟踪任务。适用于 macOS、Windows、Linux 的免费 MIT 许可证桌面应用，并提供 iOS...

测试发现：Nimbalyst monorepo installs, builds (runtime, extension-sdk, collab-bundle), and passes its full vitest unit test suite with 1546 passing test files and 12983 passing tests. 41 分钟。

极简 AI coding agent（约 1,000 行 Python），受 Claude Code 启发。可与任何 LLM 配合使用。可以将其视为 coding agents 领域的 NanoGPT。前身为 NanoCoder。

测试发现：CoreCoder installed in a virtual environment; 185 tests all green; CLI --help and --demo mode both work; ruff reports one pre-existing EXE001 lint on examples/plan_hooks_demo.py. 1 分钟。

Ovis（原 herdr web UI）：免费、开源的浏览器和手机客户端，用于 herdr。支持 Claude Code、Codex 及所有 agent 面板的聊天和实时终端，通过 SSH 连接远程 PC，并支持 Web 推送提醒。

测试发现：Ovis builds, typechecks, and passes all 6161 unit tests. The server starts and listens on its configured port. No herdr daemon is available in this container, so integration tests requiring a live herdr session were not run. 4 分钟。

响应你、你的模型、你的机器和你的规则的 coding agent。

测试发现：Zero v0.9.0 builds from source, passes all 87 test packages, passes go vet, format check, vulncheck, smoke test, and answers all CLI commands (help, models list, providers list, doctor) on linux/amd64. 9 分钟。

响应你、你的模型、你的机器、你的规则的 coding agent。

测试发现：Zero builds from source, all 87 test targets pass, release build produces version 0.9.0 and passes smoke test, and the CLI launches and responds to commands. 9 分钟。

Spec-Driven Development 结合组织治理。Specs 告诉 AI agents 要构建什么；Charter 管理它们如何构建。Git 原生任务、可强制执行的工作流，以及跨 AI coding agents 和 harnesses 的一致工程标准。

测试发现：Spec Kitty CLI 3.2.7rc1 installed in venv at /work/repo/.venv, all major test suites pass with 1903+ tests passing and 0 failures, spec-kitty init creates a working project with Codex agent integration. 82 分钟。

Agent Skills 教授 coding agents 编写、审查和移植地道 Rust 代码，并附带经过测试的 clippy 配置。适用于 Claude Code、Codex、Copilot、Cursor 和 Gemini CLI。

测试发现：All 3 BeautifulRust skills install via the README command into ./agents/skills with a skills-lock.json and list with source morganlinton/BeautifulRust; python3 scripts/check-skill.py and the 25 unittest checks pass with 0 errors; scripts/check-lints.sh reports all 28 expected lints fired once a local rustup clippy... 7 分钟。

Pi coding agent 的 Token-efficient MCP adapter

测试发现：Installed and running on Node 22.23.3. Tests pass 1997/2000, typecheck passes, public exports verified, CLI works. 9 分钟。

Ouroboros, self-creating AI agent. Born Feb 16, 2026.

测试发现：Ouroboros 7.4.4 installed and builds; CLI launches, web server serves HTML UI on port 7777 (HTTP 200), default-lane pytest suite passes. 54 分钟。

Gentle Shell 是一个 Pi 原生的 AI 辅助开发工作区，将并行 coding agent、实时任务、Git 差异、持久记忆和权限控制整合到一个终端界面中。由 Gentle AI 和 Organic Driven Development (ODD) 驱动。

测试发现：pnpm install succeeded, all 3981 unit tests pass (0 fail), provider-contract check passes, runtime-harness passes, and gentle-shell --version reports gentle-shell 3.7.0 / pi 0.87.1. 42 分钟。

Reason Network: 你的团队的 coding agents，互相交流。`reason network` 背后的引擎。

测试发现：The reason-network Rust project builds, passes all 29 unit tests, passes clippy with -D warnings, and passes all 3 e2e test suites (hooks 43/43, guided 6/6, two_machines 49/49). 10 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。

让 AI coding agents 具备架构感知能力：基线优先、证据验证、漂移检查，并在长任务中保持安全。

测试发现：The Aegis method-pack verification suite passes fully: all 38 layer1-fast-check tests, 2 e2e bootstrap checks, 275/284 unit tests, and the doctor check all report green with zero infrastructure failures. 15 分钟。

Claude Code 记忆、Codex 记忆、Cursor 记忆、OpenCode 记忆、Gemini CLI 记忆、Windsurf 记忆和 Cline 记忆，覆盖 44 个 agent。AI 编程记忆、Claude Code 历史、Cursor 历史、本地 coding agent 记忆、MCP 记忆服务器和...

测试发现：The repository builds with Go 1.25.5, the full Go test suite and the extensions/dsh npm suite pass, the built binary answers version/help and runs a real index, search, files, and deepseek plugin install on fresh state, and three pre-existing test bugs (imported-scope `files`, ESM dsh plugins, tool-fingerprint... 59 分钟。

本地优先的桌面伴侣平台，支持动画宠物、插件 SDK 和 coding-agent 集成。

测试发现：pnpm install, build, test (desktop + plugins), and check all pass, and the Electron desktop app launches on a virtual display showing a pet window with reactive idle animations. 58 分钟。

用于 Claude Code、Codex、Cursor 和其他 coding agent 的 Agent 技能与专家团队。加载一个技能，或安装领域预设、项目钩子和带测试门的并行工作流。

测试发现：oh-my-agent v13.2.1 builds successfully, the oma CLI binary prints version 13.2.1, and all 4264 tests in 323 test files pass with no failures. 53 分钟。

cloudroom 是一个开源平台，可高效运行 1,000 个以上的 agent

测试发现：Cloudroom core built from source, passes fmt/clippy and all 31 Rust unit/integration tests, boots as a real HTTP service on loopback, and completed an authenticated session (start, prompt, recorded history, graceful restart/resume) against a local PostgreSQL 16 mock with both schema migrations applied. 33 分钟。

实体级 git 合并驱动。解决 git 在 independent agents 编辑同一文件时产生的虚假冲突。与基于行的合并相比，减少约 95% 的冲突。

测试发现：The full Rust workspace builds from source with all 6 crate members (weave-core, weave-crdt, weave-driver, weave-cli, weave-mcp, weave-github). All 505 tests pass (0 failures). The weave CLI binary at target/release/weave responds to --version (0.5.4) and --help, the preview/check/explain/setup subcommands all accept... 13 分钟。

通过 Telegram 的 OpenCode 移动客户端：在手机上运行和监控 AI 编码任务，所有内容都在本地机器上运行。支持 OpenCode V2。

测试发现：The project installs, builds, typechecks, and lints cleanly with Node.js 22.14.0. The test suite passes 1969 of 1980 tests. All code-related tests pass, the 4 remaining failures are in e2e proxy connectivity tests that require IPv6 loopback, which is unavailable in this container. 16 分钟。

用于多 agent 软件开发的本地开源工作台。

测试发现：Helmor frontend builds (vite success), sidecar tests pass (418/419), frontend tests pass (883/883 across 87 files), Rust library tests pass (1557/1558 with 1 pre-existing env-var-comparison failure). 40 分钟。

面向 Claude Code、Codex、Cursor 和 Gemini CLI 的平实语言 agent 技能：诚实的会话报告、直接的结论、可遵循的步骤。MIT 许可证。

测试发现：Open Steps skill pack is installed for Codex CLI: all 8 skills copied to the shared skills folder, the routing block is in ~/.codex/AGENTS.md, both session-start and stop hooks are wired in ~/.codex/config.toml pointing at the correct files, doctor.sh reports no faults, and both the hook test suite and eval test suite... 5 分钟。

将任何 repo 变成适用于 Claude Code、Codex、Cursor 和其他 coding agent 的 agent-ready 工作区。

测试发现：harness 0.1.10 Rust binary builds, passes all 30 tests, installs the full core (27 files) into a fresh directory, and passes doctor and status checks. 46 分钟。

一个角色，多个编码 agent。通过本地 Web 控制台，跨 codex、claude、opencode、grok、omp、hermes、pi、zcode 替换或注入系统提示。

测试发现：agent-purge CLI is functional from source checkout: all 8 subcommands (help, status, ls, use, mode, matrix, detect, web) work correctly, web server serves UI on port 7788 returning HTTP 200, and 70 test assertions pass. Added missing neutral.md persona template so first-run persona auto-generation works. 4 分钟。

为 coding agents 提供即时回忆。搜索你机器上已有的历史记录。相当于 Git blame，但针对 agent sessions。

测试发现：ctx 2.2.8 CLI binary installed and verified: version command succeeds, code graph indexes the entire repo with 97k+ nodes and 167k+ edges, graph search resolves symbols with call/import chains, sift compresses command output, and all CLI subcommands (status, doctor, docs, blame, setup, daemon, upgrade, integrations)... 4 分钟。

PiDeck 是一个开源的桌面工作台，用于在本地项目目录中统一管理 PI | DSH Agent 会话，并支持导入 Codex 和 Claude 本地会话以便统一浏览和恢复。支持多项目工作区、会话历史、Git 集成、内置终端、模型配置和插件管理，基于 Electron 构建。

测试发现：PiDeck v0.7.7 builds from source, its 220 tests pass, the Electron app launches on Xvfb with full subsystem initialization, and the packaged linux-unpacked binary (301 MB) is present at release/linux-unpacked/. 16 分钟。

面向 Agents 的 AWS 技能

测试发现：The AWS Agent Skills plugin installs and is enabled via codex plugin add, the check-aws-updates.py script ran successfully against live AWS RSS feeds and found 8 significant updates, and a codex exec invocation accessed the iam skill file and answered from it. 5 分钟。

你的 AI 会遗忘，而这个会记住。面向 vibe 编码者、产品负责人、CEO 和真正的构建者的规格驱动编码 harness：自我改进的上下文记忆、15 个 agents、33 项技能，配合 /goal、agent-team 和工作流在自动运行循环中工作，0...

测试发现：Node 22 installed, resolve-manifest.js fixed (exclude bug), e2e assertions widened: all 4 e2e tests pass, install succeeds end-to-end with discover-skills OK, compute-sync-plan reports full sync (0 stale/0 missing). 28 分钟。

在终端或浏览器中，以实时流程图的形式观看 Claude Code 或 Codex 会话。

测试发现：Zoetrope builds from source, passes all 216 tests, passes clippy and fmt checks, and the zoe binary reads and inspects both Claude and Codex session transcripts, outputting complete agent trees with tool counts and metadata. 4 分钟。

通过从零构建 Claude Code 来学习 harness engineering。免费开源课程：8 篇文章、6 个视频、1 个代码库。

测试发现：Installed and built with uv, 2978 unit tests and 80 integration tests pass, format and lint clean, decode CLI reports version 0.1.0. 5 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。

同时运行 Claude Code、Codex 和 Gemini，每个都在自己的 git worktree 中

测试发现：On this container, with user-local Node 22.23.3: npm install succeeds, 4818 tests pass, npm run check is clean, the full build produces an AppImage and a .deb, and the packaged desktop app launches on the Xvfb display and keeps running while its renderer processes IPC. 17 分钟。

在 Claude Code 上使用 Obsidian + Graphify，每次会话的 token 数量减少多达 71.5 倍。包含持久化记忆、代码库知识图谱和聊天导入管道。包含 PT-BR。

测试发现：All 6 CI checks pass (py_compile, ruff, JSON block validation, README parity, shellcheck), both Python scripts process real data producing correct Obsidian markdown with YAML frontmatter and auto-tags, the shell sync wrapper orchestrates the full pipeline, and the optional claude-conversation-extractor tool is... 5 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。

生产就绪的开源终端 coding agent，具有可读、分层的代码：权限规则、操作系统沙箱、MCP、技能、子 agent，以及 Anthropic、OpenAI 兼容、Gemini 或本地模型。

测试发现：eagent 0.1.1 installs, builds, runs (dist/eagent.js --version prints eagent 0.1.1), and passes all 68 production verification tests covering CLI, tools, MCP, sandbox, skills, agents, hooks, plugins, UI rendering, and resilience. 6 分钟。

CodeAgent 是一个开源的 coding agent，常驻在你的终端中。它理解你的仓库，规划多步骤变更，编写和编辑文件，运行命令和测试，并持续迭代直到任务完成，整个过程通过透明、可扩展的工具循环实现。

测试发现：CodeAgent v0.1.0 is installed with all 41 Python dependencies in .venv, the 67-item verify regression suite passes, and the app launches showing its banner and version. 6 分钟。

面向 AI coding agents 的代码库 wiki。记录代码无法表达的内容：决策、流程、不变量、注意事项。

测试发现：The codealmanac package is installed and functional: CLI responds to --help and --version, all 564 pytest tests pass, ruff lint passes, and doctor reports a healthy install with 71 indexed wiki pages. 8 分钟。

你的 AI agents 的任务控制

测试发现：luvus 0.14.2 builds successfully, the full locked test suite passes (2198/2198), the binary responds to --help, --version, doctor, and server status commands. One test assertion was relaxed to account for bracketed-paste wrapping in PTY panes, and Node 22 was installed for OMP extension TypeScript validation. 24 分钟。

Chrome 侧边栏扩展，让 DeepSeek Harness 直接操作您的浏览器，无需视觉能力。

测试发现：The dsh-browser workspace pnpm-installs, builds, typechecks, and passes all 528 unit tests plus the real-DSH smoke test; the dsh web server starts, serves the UI at 200, exposes the bridge-config endpoint, authenticates via token→cookie exchange, and responds with correct RPC results via both the HTTP /api carrier and... 28 分钟。

AI 编码 agent 的本地可观测性。Agenttrail Map 展示项目结构和活动；Agenttrail Kitchen 以 3D 方式可视化任务和角色贡献。

测试发现：All 71 Kitchen tests pass, the server launches on 127.0.0.1 and responds with HTTP 200 on / and /api/state, with the project watched and Codex discovery active. 3 分钟。

开源、自托管的 AI 应用构建器：agent 在隔离沙箱中构建真实应用，运行于你自己的服务器上，每个应用都有预览 URL。一条命令即可自托管。MIT。

测试发现：Go 1.22.7 core test suite (23 packages) and console React/Vite unit test suite (33 tests) both pass; make build compiles sandboxd/runtimed Go binaries and bundles the console SPA; gofmt and go vet report no issues. 12 分钟。

在终端窗格中查看 agent 的 diff，并将评论发送回 Claude Code、Codex、OpenCode 或 Pi。这是一个 herdr 插件。

测试发现：All 1009 tests pass across 8 test suites, fmt-check and clippy pass, release binary builds at 17MB, and the TUI launches via PTY rendering its full interface. 10 分钟。

适用于 Pi、OMP 和 Prime Agent 的桌面工作区

测试发现：GooeyPi dependencies install cleanly, all 1927 unit tests pass, and the Electron app launches on the virtual display. 3 分钟。

Agent Plugins for AWS 为 AI coding agents 提供技能，帮助你在 AWS 上进行架构、部署和运维。

测试发现：All 125 pytest tests pass, all linters pass (markdownlint, ajv manifest validation, cross-reference validation), dprint formatting check passes, and reference validation passes (8 pre-existing orphaned files, 0 broken links). The full `mise run build` completes lint/format/validate/test stages successfully; security... 7 分钟。

Pi Coding Agent 的 Web UI，可在真实工作区中保持会话活跃。

测试发现：PI WEB installs to npm and builds (npm run verify and npm run build exit 0); the session daemon and web server run and serve 200 on /api/pi-web/version and /api/pi-web/health, and a real Pi session started through the API, prompted through the app stack, and persisted an assistant answer via the mock OpenRouter... 28 分钟。

开源跨 agent 记忆层，通过 MCP 为编码 agent 提供支持。兼容 Claude Code、Codex、Cursor、Windsurf、Gemini CLI、Antigravity、OpenClaw、Hermes Agent、Oh-my-Pi、Pi、Copilot、Kiro、OpenCode 和 Trae。

测试发现：Memorix builds and runs from source in this container: npm install, npm run build and lint exit 0, the full vitest suite passes 3177 tests, the CLI stores and retrieves observations and prints version 1.9.6, and both the stdio MCP smoke and the stateless HTTP MCP smoke complete successfully. 11 分钟。

为 Claude 和其他 coding agent 提供的动画化、可探索的图表，基于简短的 JSON 规范。

测试发现：SeeCode v0.6.1 renders all 45 diagram types from JSON specs, exports to SVG, imports from 20+ diagram formats, and passes 215/226 tests with 11 browser-dependent tests skipped. 5 分钟。

从 Obsidian 中浏览、创建和编辑 AI agent 文件，支持 Claude Code、Cursor、Codex 及 12 种编码工具。

测试发现：In this container, the repo's full check gate passes: dependencies install cleanly with the frozen lockfile, the audit is clean, lint has 0 errors, all 24 tests pass, and the production esbuild build completes with no errors. 5 分钟。

Git 原生的持久化内存，专为 AI coding agent 设计。实现了 Google OKF v0.2，具备亚 300µs 的内存 BM25 搜索、嵌入式 MCP 服务器和渐进式披露。将 token 膨胀削减 80%，无需任何外部数据库或依赖项。纯 Go 构建。

测试发现：The Go toolchain (v1.27.2) and bin/okf binary are built and functional: all 7 Go test packages pass, knowledge/ bundle validates with 21 concepts and 0 errors, bootstrap scaffolding creates valid starter packs, and search/show commands return correct results. 9 分钟。

停止重新教导你的 agent。让你的 agent 的记忆像大脑一样工作。Hippo 是 coding agent 的长期记忆。它是你的 AI harness 的关键层，连接你的不同工具（Cursor、Claude Code、Codex）。它将你的专有数据完全保留在本地，并且会随着时间推移真正学习。

测试发现：On Node 22.16.0 (user-space install), hippo-memory installs, builds (tsc + UI), passes 4212 of 4234 tests, and its CLI, MCP server, HTTP server, and packed-tarball smoke test all run successfully in this container. 7 分钟。

开源 TypeScript 终端 coding agent，专为 DeepSeek-V4 设计，基于 DeepSeek 强劲的性价比和超低缓存定价，通过工程实现字节稳定前缀和缓存复用 fork，使跨会话记忆与连续自纠错层几乎不增加 token 成本；支持 1M 上下文、Skills/MCP/Hooks，兼容...

测试发现：Build succeeds via tsc, all 1832 tests pass across 191 test files, and the compiled binary runs and prints version dao-code v0.5.9. The app requires a DeepSeek API key for real usage, which is expected and documented behavior. 11 分钟。

一个持久化的、多人的、移动优先的 Web 应用，专为 Pi agents 设计，基于 Pi Durable 构建。

测试发现：Pi Pocket installs with Node 22.23.3, passes npm check/lint/format:check and the full test suite (412 passed, 0 failed), and serves on its launcher at 127.0.0.1:8899 with working sign-in (200 app root, 303 on login, 401 on bad token). 6 分钟。

面向 AI coding agents 的本地代码智能 MCP 服务器与 CLI

测试发现：codanna 0.16.0 builds from source, indexes code locally with semantic search using a downloaded local embedding model, serves MCP over stdio/HTTP with health OK, and runs 1706 of 1730 total tests (24 watcher tests fail due to container inotify limits). 27 分钟。

一个包含 Pi Coding Agent 扩展的 monorepo

测试发现：npm install succeeds, npm test passes 5684/5686 tests, npm run check passes all typechecks and boundary checks, npm run format is clean, all on Node 22.23.3 with vitest 5.0.4. 50 分钟。

查看你的 coding agent 做了什么以及花费了多少。将每个任务分解为工作步骤：使用的工具、更改的文件、运行的测试、花费的时间和 token。本地优先的仪表板，适用于 Claude Code、Codex、OpenCode 等。无需登录，无需遥测。

测试发现：agentacct 0.12.11 installed in ~/.agentacct venv, all 4888 tests pass, demo command produces run reports, local JSON API returns 200 on /. 50 分钟。

面向编码 agent 的持久化项目知识图谱。提供语义搜索、进程内嵌入和 Web 浏览器的 MCP 服务器。

测试发现：MegaMemory v1.6.2 builds, all 153 tests pass, the MCP stdio server responds to all 9 tool calls, and the web graph explorer serves HTTP 200. 2 分钟。

在 MCP 客户端上的 Shell 和 coding agent

测试发现：wcgw 5.6.5 is fully installed, all 63 pytest tests pass, the MCP server entry point (wcgw) and CLI (wcgw_local) both respond to --help, and all package imports resolve cleanly. 2 分钟。

面向 AI coding agent 的上下文压缩：在工具输出进入模型之前进行压缩，将重复内容去重为 13-token 引用。支持 Claude Code、Cursor、Codex、Kiro、Zed 以及任意 MCP 客户端。Rust 实现，零 LLM 调用。

测试发现：The sqz workspace builds and its 1715 tests pass; sqz 2.1.1 runs, compresses with byte-exact expand, the sqz-mcp stdio server completes the MCP initialize handshake and lists 7 tools, and the local proxy answered HTTP 200 on POST /v1/chat/completions with a compressed body. 13 分钟。

面向真实产品的自主开发。受管控的 agent 根据你批准的规范进行构建，遵循你设定的规则，并且在另一个 agent 验证通过之前不会合并任何内容。你的决策、判断和品味是工作必须达到的标准。宣告你的愿景，让 agent 去构建它。

测试发现：Pullboard 0.9.0 is installed, the CLI answers commands, the view server binds a port, the spec check reports 0 errors, and the test suite passes 1410 of 1414 tests (1 container-specific parent-death runner test fails, 3 skip without Chrome). 59 分钟。

查看你机器上每个 AI coding agent 正在做什么，以及其背后的 API 是否是你付费的那个。Claude Code、Codex、Cursor、Gemini CLI 等：一个端点覆盖 44 个模型供应商，A, F API 真实性评级，按 agent 的权限、使用情况和提示记录。

测试发现：Server binary installed via curl|bash and starts on localhost:17000; sign-in (admin/123), API endpoints, relay token, provider CRUD, and LLM relay (non-stream and streaming) all verified end-to-end with a mock OpenAI upstream returning HTTP 200 with correct response body. 10 分钟。

面向开放模型的开源软件工厂

测试发现：codeaf v0.8.0 release binary installed, builds from source with SHELL=/bin/bash, all 31 law/test-quick packages pass, real model call to DeepSeek returns correct output. 31 分钟。

在同一个运行时后面运行 Claude Code、Codex、Antigravity、Grok Build 和 OpenCode，并且只有当它未编写的检查通过时，一次运行才算完成。跨引擎扇出、委员会、会话交接、支出上限、MCP 服务器、ACP agent。

测试发现：Project builds with npm run build, all 1724 vitest tests pass, CLI binary runs and reports version 7.5.5 with all 15 commands listed. 4 分钟。

MCP 工具：让你指向 DOM 元素，供你最喜欢的 agent 编码工具使用。让 AI 看到你所看到的。

测试发现：Full monorepo builds clean. Server package (23/23 tests passing) and Chrome extension package both build. CLI binary starts, prints leader banner, and accepts WebSocket connections. 2 分钟。

一个持久化的、基于文件的记忆层，为编码 agent（如 Claude Code、Codex 等）赋予项目大脑（持久化的决策、需求、约束），通过零依赖的 CLI 实现。

测试发现：All 50 project tests pass. Every CLI subcommand runs correctly: init scaffolds BRAIN.md and brain directory, CRUD page operations work end-to-end, hooks install/uninstall cleanly, and wiring creates agent config files. Zero external dependencies or credentials required. 4 分钟。

使用单个命令，在你的 PC 上或云端，在并行的沙箱虚拟机中运行多个 agent。

测试发现：The AgentBox monorepo builds and all 46 package test suites pass (0 failures, 0 cached failures), the agentbox CLI binary responds with correct version/help output, and rsync is available at /home/runner/.npm-global/bin/rsync for the sync-layer tests. 19 分钟。

规范驱动开发与智能压缩，结合 Ralph Wiggum 循环和结构化规范工作流的 Claude Code 插件。

测试发现：Smart Ralph plugin suite fully passing: 342 Bats tests and 13 Python tests all green, with stop-watcher, state management, integration, prototype phases, interview framework, phase gates, workflow guardrails, and platform validation all verified. 23 分钟。

开源设计 MCP 服务器 + shadcn/ui 注册表：适用于 Claude Code、Cursor、Codex、GitHub Copilot 及任何 MCP 客户端的 AI 设计系统。包含 31...

测试发现：The registry repo builds and passes its suite (22 tests, typecheck clean, generator idempotent), and the published better-design CLI is installed with a live free API key that the hosted MCP server accepted, with the MCP server wired into ~/.codex/config.toml. 3 分钟。

AppWorld：一个可控的应用与人物世界，用于 Benchmarking 函数调用和交互式 Coding Agent，ACL'24 最佳资源论文。

测试发现：AppWorld installs and builds; all 1851 package tests pass; the CLI (appworld --version prints 0.2.0.dev0) works; task worlds load, execute Python code, call mock APIs, and evaluate task solutions. 37 分钟。

Codex/Claude Code/Cursor 的开源增强版，专注一句话实现复杂长程任务（教育、编程、办公、生活、娱乐、游戏）。部署在你自己的服务器上，团队用浏览器打开就能编程，包括手机。CLI & Web UI 双入口，Multi-Agent 协作，原生直连千问/DeepSeek...

测试发现：Python FastAPI service runs on port 8000 with health endpoint OK; Java Spring Boot backend runs on port 8080 with actuator health status UP and SQLite global+project databases operational; React frontend builds cleanly with 207/223 tests passing; all three tiers (Python 105/107, Java 2392/2454, React 207/223) pass... 18 分钟。

MiniMax Code 团队用于构建 MiniMax Code 的工作流程和技能。开源，安装并使用。

测试发现：Both ronny and jiachen stacks install correctly as Codex plugins with all skills present, and pass both the repository's structural checks and the official MiniMax Code Plugin V1 validation. 3 分钟。

为你的 coding agent 提供代码图谱和记忆，集成在一个 Rust 程序中。适用于 Claude Code、Codex、Cursor 以及任何 MCP client。

测试发现：Graphicle 0.1.5 installed via prebuilt binary, setup completed successfully on the repository (code graph built, agent connected, hooks installed, web app serving HTTP 200), all core commands function without errors. 4 分钟。

Git worktree 管理器，带模糊查找功能 - 可同时处理多个分支，非常适合并行 AI coding 工作流

测试发现：The gwq binary builds and installs with a downloaded Go toolchain, runs its full test suite (go test ./... exits 0 with all ok), and its add/list/get/status/exec/remove commands operate on real git worktrees as confirmed by git worktree list and command output. 4 分钟。

IDE，一个具有原生体验的 AI 编码工作区，将聊天、规划、agent 执行和项目导航融合为统一的桌面体验。

测试发现：npm install completes, full build (typecheck + esbuild + vite) succeeds, all 117 test files (709 tests) pass, and the Electron app starts with the dist bundle rendering the agent window via did-finish-load. 8 分钟。

`agent-spec` 是一个 AI 原生的 BDD/spec 验证工具，用于任务执行。

测试发现：agent-spec 1.4.0 builds and all 1111 tests pass on Rust 1.98.1 with no modifications. 3 分钟。

一个基于 AI 的 GitHub 代码审查工具，利用 LLM 检测高置信度、高影响的问题，例如安全漏洞、错误和维护性问题。

测试发现：All 157 tests pass and gito CLI version command runs from a Python 3.12 venv, with only the version-shell test requiring a PATH fix. 3 分钟。

ADE（Agentic Development Environment）是一个面向 AI coding agents 的规范驱动环境，让你的规划成为持久、共享的项目上下文。

测试发现：All 1171 tests pass. The Electron app launches on the virtual display (Xvfb :99), creates its window menu, and runs without crashing until terminated. 12 分钟。

终端 agent 协同工作的地方。Claude Code、Codex、Copilot CLI 和 pi 共同处理同一任务，在同一使命中，每个 agent 在真实终端中保持自己的 TUI。

测试发现：The Rust workspace builds successfully for all Linux-compatible crates (runner-core, runner-backend, runner-cli, runner-terminal). All 1318+ tests pass. The runner CLI binary outputs version 0.12.9 and responds to all subcommands. 16 分钟。

面向 AI agent 的 ADE。自动化其余一切。做你真正的工作。

用于导入、管理、运行和注入 APM 支持的助手包的本地编辑器。

测试发现：APM Studio installs, passes all 915 tests, builds client and server, and serves HTTP 200 on its API and frontend endpoints. 2 分钟。

用于 agent 的受限自主插件 harness。从意图到实现：就绪，然后开始。

测试发现：Repository builds all 5 platform targets (Claude, Codex, Grok, Agent Plugins, Antigravity), passes full verification suite, and the Codex plugin installs and enables correctly in the Codex CLI. 3 分钟。

为 Codex、Claude Code、OpenCode、PI 等提供原生会话控制平面。可在多台机器间运行、恢复和移交编码会话。

测试发现：Bridge server launches, listens on configured port, authenticates via Basic Auth, and responds to API routes; all 600 bridge tests pass with zero failures. 4 分钟。

掌握 OpenCode，这个开源的 AI coding agent：设置、agents、技能、插件、MCP、Zen 与无头 CI。

测试发现：All 17 markdown files pass markdownlint, codespell, and link checks; opencode.json and tui.json are valid JSON; the repo is clean with no working tree changes. 3 分钟。

运行 coding agents 而无需像保姆一样照看它们。让任务从开始到结束保持专注、有边界、可检查、可问责。最终，让你的 agents swarms 整夜运行，拿回你的时间。

测试发现：MartinLoop 0.6.9 builds from source, the martin CLI launches with all subcommands, the SDK imports with all exports, 6 vitest packages pass all tests, and 166/166 script-level release tests pass on Node 20.20.2 with pnpm 10.33.0. 17 分钟。

vi 编程 Agent 桌面应用：流式时间线、Git 审查（带 hunk 暂存）、会话树、vi 扩展的原生 UI。支持 Windows、macOS、Linux。

测试发现：pi Desktop 0.6.0 builds from source, all 1384 unit tests pass, all 275 contract tests pass, typecheck passes, and the Electron app launches on the virtual display. 5 分钟。

开源、设备端的语音+视觉层，专为 AI agents 设计。可接入任何模型或 coding agent；整个语音循环（VAD、STT、TTS、barge-in）均在本地运行。是 ElevenLabs Agents、Gemini Live 和 OpenAI Realtime 的开源替代方案。

测试发现：The OpenLive monorepo installs, builds, type-checks, runs all 82 tests successfully, and serves the web UI on port 3000 with the agent service on port 8787, both returning HTTP 200. 4 分钟。

Pi Desktop: 一个免费、开源的桌面 GUI，用于 Pi coding agent 和 oh-my-pi。可在同一窗口中聊天、编辑文件、运行终端和查看差异。支持 Linux、macOS 和 Windows。

测试发现：Pi Desktop v0.1.9-alpha builds from source, passes all 1246 tests, and renders its GUI correctly on a virtual display. 5 分钟。

面向 Claude Code 和 agentic CLI 的模块化插件市场，提供经过验证的、规范驱动的技能、代理、命令和工作流，适用于 Java、TypeScript、Python、PHP、AWS 和 AI。

测试发现：All 211 tests pass, the validator CLI validates all 744 files without errors, and the Codex CLI install target copies 151 skills and generates an AGENTS.md index file. 4 分钟。

由 DeepSeek 驱动的 coding agent。

测试发现：DSCode 0.3.6 installs from source, builds via pnpm build (TypeScript compilation succeeds), all 206 tests pass (4 platform-specific tests skipped on Linux), and the CLI launches reporting version 0.3.6 with a full usage page. 7 分钟。

在一个本地优先的工作区中编排一组专家 AI agents，每个角色都使用其最佳模型，并具备可累积的记忆与技能，免费且采用 MIT 许可证。

测试发现：Daemon builds and serves health endpoint at 127.0.0.1:8080; all 16 Go test packages and all 341 frontend vitest tests pass with Node 22.23.4 and Go 1.26.1. 8 分钟。

Kelos - 用于编排自主 AI coding agents 的 Kubernetes 原生框架。

测试发现：All unit tests (45 packages) and integration tests (173 specs) pass, the CLI binary builds and answers commands, and the controller, webhook, and install CRDs work correctly against a real Kubernetes API server. 11 分钟。

ACRYL - Agent 上下文中继产生生命周期。一个持久的工作区，一个规范的上下文，任何 coding agent。

测试发现：Node.js 22.19.0 works with Corepack 0.34.0 (pnpm 11.11.0); 1263 packages installed; 24 workspace packages build, typecheck, and test; the ACRYL CLI help command prints the full interface; the acryl-web --help command works; all headless test suites pass (acryl-control, blends-core, app-persistence... 14 分钟。

将你的 coding agent 变成视频工作室：用自然语言描述视频，你的 agent 会编写时间线并生成文件。

测试发现：The repository installs, builds, typechecks, benchmarks, and passes all 282 unit tests plus 47 E2E tests including a real three-scene video production pipeline that composes HTML motion graphics, renders via HyperFrames 0.7.60 with chrome-headless-shell, burns captions with ffmpeg, and passes the delivery... 12 分钟。

面向从经验中学习的 agent 的开源基础设施。捕获生产轨迹，将经验转化为可复用的知识和技能。

测试发现：All four test suites pass (CLI 259, frontend 336, www 14, backend 184), both Next.js apps build, and ruff lint/format passes. 24 分钟。

为 AI coding agents 精选的生产级技能。为认真使用 AI 的开发者打造的经过实战检验的工作流。

测试发现：uv sync installs all dependencies, 369 repo-level and 348 package-level tests pass, all 145 packages install to ~/.claude/, eval validation and search scripts operate correctly, templates are in sync. 3 分钟。

agent 技能的任务控制

测试发现：LazySkills binary builds from source, passes its full test suite (including race detector), passes all CI checks (gofmt, vet, deadcode), and its non-TUI commands (doctor, find, update, version) execute correctly against both the filesystem and the external skills.sh registry API. 3 分钟。

为开发者和 coding agent 提供一个可靠的编辑循环

测试发现：All 931 unit tests pass, both funzzy and fzz debug/release binaries build and run correctly, config validation and schema output work as documented. 9 分钟。

在 AgentFS 内运行 Claude Code/Codex，由 LlamaIndex Workflows 编排

测试发现：TypeScript builds cleanly, all 10 tests pass, and the MCP server starts and serves all 5 filesystem tools (read_file, write_file, edit_file, file_exists, list_files) over HTTP on port 3000. 6 分钟。

任意 Agent × 任意 Provider × 任意位置。将 Cursor、Claude Code、OpenCode 或 Codex 连接到 Slack、Telegram、Discord、Feishu、DingTalk、WeCom、WeChat，并支持任意 LLM provider。

测试发现：pnpm install succeeds, tsc build succeeds, all 1336 unit tests pass, the CLI binary prints help and version, the HTTP serve mode listens on a port and answers /health with 200, and the gateway mode starts with all channel stubs and answers /health with 200. 2 分钟。

Packmind 无缝捕获你的工程手册，并将其转化为 AI 上下文、护栏和治理。

测试发现：pnpm install completed, tsconfig.base.effective.json generated, all 22 backend-test projects pass (694 suites, 15067 tests), frontend passes (165 suites, 2532 tests), packmind-cli builds and runs (version 0.36.1-next), eslint lint passes on all checked projects. 10 分钟。

一个本地优先、跨平台的 Electron 桌面工作区，用于 Pi Coding Agent，具有会话、项目文件、浏览器工具、技能、插件和消息集成功能。

测试发现：Pi Agent Desktop v0.3.0 builds, all 1106 tests pass, and Electron smoke test verifies agent-host (pi=0.87.1) with all 12 function domains (renderer, RPC, session, worktree, git, watch, download, skills, channels, chat-appearance, toolchain) passing. 11 分钟。

AI 编写代码的黑盒。被动决策图将每次 commit 与背后的 agent prompt 关联起来：漂移检测、agent 读取内容的 DLP 审计、证据驱动的风险、用于编码 agent 的 MCP 内存、签名的来源证明。一条命令，默认离线。

测试发现：The brain0 monorepo (14 Rust crates, 5 TS packages) builds and all tests pass; the HTTP server serves the GUI and graph/debug APIs correctly. 4 分钟。

面向 AI 编码 agent 的 RFC 驱动开发：想法或现有代码库 → 已验证的 PRD → 功能 → 规则 → 有序的 RFC → 已审查的代码。适用于 Claude Code、Codex、Copilot、Cursor、Gemini CLI、OpenCode、Devin 的 Agent Skills。

测试发现：Repository builds all 11 Agent Skills from prompt sources, installs into .claude/skills and .agents/skills without errors, passes all 18 checks across three test suites, and versions are consistent at 3.0.0 across install.sh, plugin.json, and CHANGELOG.md. 2 分钟。

Compose Skill (Compose Kit)：面向 Jetpack Compose 和 Compose Multiplatform 的 agent 技能。让 Claude Code、Codex、Cursor、Copilot、Gemini CLI 和 OpenCode 编写更好的 Compose...

测试发现：All 90 guard tests pass: 11 shell-guard scripts validate MVI/Koin/Nav3 architecture conventions on good fixtures and correctly reject bad fixtures, frontmatter schema validation passes on all 7 skills, the new-feature scaffold builds correctly and passes all checks after SEAM markers are implemented, install-guards.sh... 1 分钟。

用于更清晰的 coding agent 响应的确定性文本检查

测试发现：nopus installs, builds, and passes all 55 tests on Node 26 with all 5 normalized datasets verified and valid plugin artifacts produced. 3 分钟。

用于 AI coding agents 的原生策略执行层。基于 OPA/Rego 构建。

测试发现：Cupcake 0.5.2 builds from source, the CLI binary runs (cupcake --version prints 0.5.2), supports 4 harnesses (Claude Code, Cursor, Factory AI, OpenCode), compiles Rego policies into WASM via OPA, evaluates hook events against builtin policies (git_pre_check, protected_paths, git_block_no_verify), and returns... 19 分钟。

Codev 帮助人类和 agent 共同开发项目的上下文和代码。

测试发现：The Codev monorepo builds, all 6210 unit tests pass, and the CLI runs and reports version 3.3.4 with doctor verification succeeding. 13 分钟。

根据当前工作、Git 分支、终端活动和 Claude Code 会话自动命名 Herdr 标签和窗格。

测试发现：Go 1.24 downloaded from go.dev, project builds with 'go build', all 554 tests pass across 8 packages, binary exits correctly with expected error when run outside Herdr. 6 分钟。

Broccoli 将 Linear 工单转化为已交付的 PR，由 Claude 和 Codex 提供支持，并运行在您自己的 Google Cloud 或 Blaxel 容器上。

测试发现：Python 3.12 project installs via uv sync --dev, ruff check passes, pytest passes 156/171 offline (15 skipped need PG), FastAPI app serves healthz 200 and processes webhooks against a live Postgres 16 with all migrations applied. 42 分钟。

验证 Skill 内容是否符合 Agent Skill 规范，并包含额外的内容密度和质量检查。

测试发现：All 13 Go packages build and test successfully; the CLI binary runs at version v1.6.2; structure validation, link validation, content analysis, contamination analysis, and the 'check' command all function correctly against the example skill. 2 分钟。

推动 AI coding agents 养成更好习惯的自动化质量检查

测试发现：habit-hooks project installs via uv sync, pnpm install, all 939 tests pass, ruff lints clean, habit-hooks --all runs successfully on the repo itself. 18 分钟。

一份紧凑的运营章程，将 LLM coding agents 转变为自律的首席工程师：Eleven 条规则 + one 条元规则。

测试发现：The eval harness installs and builds cleanly, all 70 harness unit tests pass, all 15 fixture suites match their pinned counts, and a documented bug fix plus the --quiet task implemented over the fixture verified green test runs and correct CLI output in this container. 4 分钟。

开源、终端优先的 AI coding agent，具有完全透明的多模型路由。本地（Ollama, LM Studio, MLX, llama.cpp）或云端，你的密钥，一个 TUI。没有黑箱。

测试发现：Albatross 2.5.0 builds from source, passes all 712 unit tests, passes cargo fmt and clippy -D warnings, and can run a one-shot prompt against an Ollama-compatible API returning SSE streaming chunks. 18 分钟。

针对 Claude Code、Codex 和 GitHub Copilot CLI 的 TDD 强制执行与护栏

测试发现：@nizos/probity v1.10.1 builds, its CLI binary (dist/bin.js) prints version and help, and all 60 test files pass (550 tests passed, 26 skipped). 2 分钟。

中文优先的 Vibe Coding / AI coding 工程工作流指南：规格、agent、工作树、技能、CI 和审查。

测试发现：The static tutorial site installs with zero dependencies, serves HTTP 200 on python3 -m http.server 4173, and works in real headless Chrome: schedule/day navigation, checklist progress persisted in localStorage, and reader mode rendering the markdown course; its only 404 is the optional favicon.ico. 6 分钟。

你一直在寻找的一个仪表盘：跟踪 Claude、Cursor、OpenRouter、Copilot、Gemini、Codex 等产品的支出与用量。

测试发现：52/52 Go test packages pass (race detector on), headless subcommands daily/monthly/session/blocks/detect/version all produce correct output with real Codex CLI data, binary builds at 19MB. 6 分钟。

本地运行的 Go 静态分析引擎，内置 MCP 服务器。为 AI coding agent 提供确定性结构感知：调用图、影响分析、符号搜索等功能。

测试发现：Gograph is fully built, all 26 test packages pass, the binary reports version 1.7.5-9e4c8b4 on go1.27.0, and it successfully builds both AST and precise structural graphs of its own repository. 6 分钟。

当 5 小时或每周使用限制重置时，自动恢复 Claude Code、Codex CLI、Grok、Qwen Code、Kimi CLI、OpenCode 和 Antigravity 的会话，适用于 tmux、Zellij、VS Code 和桌面应用。

测试发现：All 1489 tests pass on Node 20.12.0; the unsnooze CLI responds to --help with full usage output. 14 分钟。

从一个统一页面管理你所有的终端和 AI coding agents。

测试发现：Go build succeeds, all 14 test packages pass, server binary CodeKanban runs and responds with 200 on /api/v1/health, and the npm package [email protected] is installed globally. 7 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。

用于与 AI coding agents 协作的安全本地开发环境

测试发现：The agentbox CLI binary is built and functional at dist/agentbox-linux-amd64, all 5 Go test packages pass, and all 195 proxy Python tests pass. 4 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。

一个用于组织 Codex 项目、工程任务、会话、Git 变更、Skills 和 MCP 配置的本地桌面控制中心。

测试发现：Frontend build and preview server work. Tauri native build and Rust tests cannot compile because system packages (libglib2.0-dev, libgtk-3-dev, libwebkit2gtk-4.0-dev) are not installed and cannot be installed without root. 3 分钟。

coding agents 的可移植、可验证工作状态

测试发现：ThreadPort 0.2.1 built from source, all 523 unit tests pass, the CLI validates schemas, renders capsules, reports diagnostics, and lists targets without errors. 4 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 Windows 桌面客户端和 GUI，适用于 DeepSeek Harness，零配置安装程序，包含 Codex、插件、技能、SSH、移动远程访问和 11 种皮肤。

测试发现：Dependencies installed, all 40 workspace projects build and typecheck. Script tests 200/200 pass. Desktop test suite 1229/1229 pass (0 fail, 26 skipped). DSH web runtime starts and serves HTTP (401 with auth active). One pre-existing issue: dsh-ssh engine.test.ts needs ssh-keygen (openssh-client) unavailable in this... 22 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。

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。

在另一个工具中恢复任何 AI 编码会话：Claude Code, Copilot, Gemini, Codex, Cursor

测试发现：Node.js upgraded to v22.14.0 (via /tmp/), pnpm installed and builds approved, project compiles with tsc with zero errors, all 28 test files pass (917/919), CLI binary functions (help, version, scan, list all work), and biome passes with warnings only. 7 分钟。

在浏览器中审查 AI coding agents 的计划、差异和运行中的应用。本地优先，可与任何 agent 配合使用。

测试发现：Built crit binary compiles, launches an HTTP server on 127.0.0.1, serves /api/health returning 200, auto-detects git changes in feature branches, serves embedded frontend, and the full test suite passes all unit and JS tests except one pre-existing intermittent flake in session lazy-threshold ordering. 42 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。

一个实时语音运行时，让 Agent 保持对话、工作和在场。面向 AI Agent 的实时语音运行时。

测试发现：The qwen-audio-agent v1.11.0 project installs, builds, launches a Gateway serving HTTP 200 on port 3101, and passes all 1789 automated tests with zero failures. 9 分钟。

一个针对较小 LLM 优化的测试框架

测试发现：little-coder v1.19.0 installs and runs: npm install succeeds (182 packages), all 715 vitest tests pass, 4/4 pytest RPC client tests pass, TypeScript typechecks cleanly, the CLI launcher starts pi and lists available models correctly. 4 分钟。

面向 AI coding agent 的非破坏性压缩网关。可将 token 开销从第 1 轮降低 25%，到长会话或饱和会话中降低超过 85%，并在相同的上下文窗口中容纳约 3 倍的轮次。由我们开源的代码原生 4B 模型驱动。即插即用，适用于 Claude Code、Cursor、Codex、OpenHands...

测试发现：Python package installed in a venv, the proxy serves on port 8088, answers /health (200) and /stats (200) and /v1/messages/count_tokens (200), and all 316 tests in the test suite pass. 8 分钟。

用于 AI coding agents 的终端会话管理器。一个 TUI，支持 Claude、Gemini、OpenCode、Codex 等。

测试发现：Agent-deck vdev builds and runs in the container. The binary responds to version, help, and ls commands. 34 of 34 directly testable internal packages pass. The WebSocket terminal bridge functions correctly. Only pre-existing environment-specific test failures: keepalive client-count assertions (container PTY behavior)... 62 分钟。

面向 VS Code 的 AI 编码 agent 梦之队。Claude Code 与 openai Codex 在头脑风暴模式下协作，辩论解决方案，并为你的代码综合出最佳方案。

测试发现：Mysti v0.4.0 installs, builds via webpack, and passes all 367 automated tests with vitest v2 on Node 18, with the vscode API mocked. 5 分钟。

一个开源图工程运行时，将编排保持在 TypeScript 中，并将语义工作委托给可替换的 Agent 运行时。

测试发现：The deer-workflow repository installs and builds cleanly with Bun 1.4.2, all 89 tests pass, and the CLI runs showing help for create/skill/run commands. 2 分钟。

面向 AI 编程 agent 的高级开发者规范。一个将未检查项视为失败的提交门控，拒绝将未完成工作标记为完成的质量控制器，以及一个针对每个逃逸 bug 的类别进行闭环修复的经验教训循环。单个 Go 二进制文件，无运行时依赖项，支持 20 多个 agent。

测试发现：Procoder builds from source with Go 1.26.7 and its full Go test suite passes (57/57 packages), procoder test reports go pass, and the procoder binary runs version/help and its gate end-to-end, flagging conflict markers in a scratch repo. 7 分钟。

用于 coding agent 的开源元 harness。定义你自己的工作流、审查者和修复循环。zeroshot 强制执行你配置的流程。

测试发现：Zeroshot installs via npm and builds from source; the CLI version, template list, UI server (200 on /ui/), and run validation all work, with 1607/1608 Rust tests passing in this container. 25 分钟。

开源 Grok Bot 替代品，自带 agent 和订阅：在您自己的电脑上，通过 iPhone、Mac、Windows、Linux 或终端，将 Claude Code、Codex、Gemini CLI、Cursor、Copilot 及 40 多个 AI agent 作为 bot 来发送消息。与 agent...

测试发现：codync-host 2.13.2 builds and serves HTTP on the configured port; all host unit/e2e tests pass; cloud and relay test suites pass fully; desktop app builds, typechecks, and passes 124/131 tests (7 SSH-related failures from missing /usr/bin/ssh in the container). 22 分钟。

Codex 内置的开源 AI 幻灯片工作室：图像原生的幻灯片，每张幻灯片都是完整的视觉画布。10+ 张高质量幻灯片，约 4 到 5 分钟完成，快速模式并行渲染每一页。实时观看整个流程：research → outline → style → render → edit → present → export...

测试发现：Web app installs, builds, serves 200 on /, /api/agents, /api/scenarios, /api/templates, creates and outlines projects via the API, and its 67-test suite passes; the Electron desktop app could not launch in the rootless container (SUID sandbox requires root). 8 分钟。

每个文件未打开。每个文件夹未探索。令牌已保存。为编码 agent 提供的本地仓库映射：用途、源码、符号和依赖项。原生 Rust CLI 和 MCP，隔离的工作树和子 agent 钩子。

测试发现：ProjectAtlas builds from source with Rust 1.98.0, all 746 unit tests pass, 113 of 115 integration tests pass, CLI binary initializes and queries the repository. 36 分钟。

用 Rust 编写的终端开源 agent。可使用任何提供商或模型。始终欢迎 Issues 和拉取请求。

测试发现：Codewhale 0.9.11 builds, passes all tests, launches CLI help/doctor/exec against a mock API. Debug profile is memory-safe; release build hits OOM on constrained hardware. D-Bus dev headers were missing and were mocked via vendored crate sources. 42 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。

每个 agent 的持久进程，内存可在重启后保留

测试发现：Node 22.19.0 with pnpm 9.15.0 installed, all dependencies resolved and built, type checks pass, boundary checks pass, 68 of 71 tests pass across packages/pi and packages/sandbox-adapter. 39 分钟。

Agent 技能，用于将架构图、流程图、时序图、数据流图和生命周期图绘制为手动放置的 SVG，并遵循统一的 lint 检查风格。

测试发现：With the patched test script, npm test passes 825/825 on the container's Node 18 and on Node 22, npm run lint:svg:all reports 11 files with 0 errors and 0 warnings, the skill installed into ~/.agents/skills/svg-diagram and runs from there, and the README's broken-SVG example reports the documented 3 errors/7 warnings... 5 分钟。

Solo Agent：一个开源、本地优先的工作空间，人类和 AI coding agents 通过频道、任务、团队和持久化记忆进行协作。

测试发现：In this container the Solo stack builds and runs on real PostgreSQL 16: server :8080 and paired daemon :8081 answer 200 on /readyz and /health, frontend :3000 serves the app, 81 migrations are applied, all Go unit tests pass, and 7 real end-to-end tests pass, while agent-powered onboarding waits on a logged-in LLM CLI... 56 分钟。

omg.dev：用于 claude、codex、cursor、opencode、pi、grok、jcode 的远程控制，带有移动客户端

测试发现：Dependencies installed, typecheck passes, the server starts on 127.0.0.1:8766 and all API endpoints respond correctly, and the test suite runs 5356 passing tests with only environment-specific failures related to Chromium, dbus, mobile native modules, and AI agent credentials. 38 分钟。

与 coding agent 协作的最佳方式。并行运行 fifty 个，让它们互相生成，并在一个地方审查所有内容。

测试发现：All headless Rust crates compile, pass clippy and fmt checks, and 114+691 tests pass; the diri-remote binary probes platform metadata and captures remote environments; the GPUI desktop app compiles through cargo check but cannot link in this 8GB container. 34 分钟。

DeepCode：开放式 agent 编码（agent 框架、循环工程与多 agent 编排）

测试发现：The Python package is installed and builds, the CLI reports version 2.1.0, the app-server runtime probe passes, and the full test suite (1586 tests) passes. 3 分钟。

免 ROOT 免 Termux，在手机上运行 DeepSeek Harness。完整 Ubuntu 环境 + proroot 零 ptrace 开销 · AI 输出实时上屏 · ADB 直连 · 数据不丢

测试发现：Test engineer successfully built JDK 17 + Android SDK 37 toolchain, compiled and confirmed all 32 pure-logic invariants across the 4 core util test classes pass without Android dependencies, but the full Gradle build is blocked by missing offline asset files that are CI-generated. 17 分钟。

AI 上下文的 ripgrep：一个零依赖的 C++23 CLI + MCP 服务器，用于编码 agent。无需阅读仓库即可找到所需内容，然后检查你构建的是否符合预期：blast radius、要运行的测试、质量差异。签名比函数体少 74.7% 的字节；每个猜测都被标记，每个损失都公布。带着地图出发。

测试发现：Build succeeds (plain dev build/ and Release build-install/), binary parses the repo source tree and emits deterministic minified XML with ranked symbols via Personalized PageRank, installed to /home/runner/.local/bin/ripwire with skills and hooks, --version/--help work, Python indexing confirms correct function... 21 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。

一个快速的 coding agent、rpc 插件、子 agent、粗略编辑和 MCP

测试发现：phi 16 MB binary builds from source, runs, and all tests pass on Go 1.26.9 linux/amd64. 2 分钟。

IDE，一个具有原生体验的 AI 编码工作空间，融合了聊天、规划、agent 执行和项目导航，提供统一的桌面体验。

测试发现：Dependencies installed; 709 of 709 tests pass; typecheck and the Vite renderer build succeed; the Async IDE desktop app starts on the virtual display, shows its main window, and keeps running past 6 minutes with auto-update disabled. 42 分钟。

JoySafeter：一个企业级 AI Agent 平台，不仅仅是聊天。构建、运行、测试和追踪自主 Agent 团队，支持可视化编排……

测试发现：Backend Python dependencies installed successfully, all 247 backend tests pass (1 expected skip). The test suite validates FastAPI routes, WebSocket handler, schema serialization, permissions, service layer, Pydantic adapter, prompt templating, and chat protocol parsing. 16 分钟。

从工单到已审阅的拉取请求。免费且开源，运行在你的机器上。

测试发现：The no-human project (v0.2.0) is fully installed: uv sync completes, web board builds from source, the nh CLI responds with version 0.2.0, nh start binds and serves HTTP 200 in setup mode, the full node test suite (1584/1584) and the Python repoguard suite (142/142) pass. Task dispatch requires a Claude subscription... 71 分钟。

一个强大的 MCP 编码工具包，提供语义检索和编辑能力，是你的 agent 的 IDE。

测试发现：Serena is installed, initialized, and the MCP server starts successfully with all 708 tests passing. 12 分钟。

面向 AI coding agents 的预执行防护，在工具调用运行之前阻止破坏性的 Git 和文件系统命令及常见的敏感文件访问尝试，支持 Amp Code，Antigravity CLI，Claude Code，Codex，Cursor，DeepSeek Harness，Devin CLI，GitHub...

测试发现：bun install and bun run build succeed. bun run check passes lint, formatting, typecheck, knip, and duplication checks, with 3673 tests passing out of 3677 total across 213 files. Coverage is 98.91%. All E2E tests pass (packed-runtime, hermes-openclaw, and protection contracts). 15 分钟。

Multi-harness 控制平面，用于 Claude Code、Codex、Cursor 和 OpenCode：跨多个 Claude/Codex 订阅的配额感知轮换、共享线程上下文和跨模型审查。

测试发现：pnpm install completed, pnpm build succeeded (31 packages in 18s), the CLI runs and reports version 3.9.8, the full test suite passes (350 test files, 4692 tests), canary stories pass (6 files, 50 tests), and typecheck passes (61 tasks). 20 分钟。

一个桌面应用，汇集你的所有 AI coding Agent：Claude Code、Codex CLI 和 Gemini CLI。自动检测、一键安装、统一聊天、文件资源管理器、终端和编辑器。本地优先。基于 Tauri + Rust + Vue 3 构建。

测试发现：Frontend builds with Vite, backend compiles with Cargo, all 98 tests (13 JS + 85 Rust) pass, and the desktop binary launches on Xvfb and starts its proxy server. 16 分钟。

CLI 代理，可将常见开发命令的 LLM token 消耗降低 60-90%。单个 Rust 二进制文件，零依赖。

测试发现：RTK (Rust Token Killer) v0.42.4 builds from source, passes all 2936 tests with 0 failures, and runs as a working CLI proxy producing token-compressed filtered output for ls, git, cargo, and other commands. 8 分钟。

CodeNomad：让 AI 编码更强大的指挥中心。

测试发现：CodeNomad monorepo installs, builds, and typechecks; the Electron native test suite runs 322 of 326 tests passing (1 zsh-missing fail, 3 skipped); the CLI server starts and serves HTTP 200 on both HTTPS and HTTP; the Playwright browser test framework runs correctly. The single failure is zsh not being present in the... 20 分钟。

将你的 coding agent 会话以地图和时间线形式回放

测试发现：Python package agentplay 0.2.2 installed from source, 991 tests pass, CLI builds sessions from Codex history and serves a functional web viewer with API returning available=true. 5 分钟。

使用 pi 进行氛围编程，在一个轻量级、真正的 IDE 中，它会根据你的工作方式自我定制：你需要的氛围

测试发现：890 deps installed, 14 packages typecheck, 14/14 test suites pass (987 in server alone), web app builds, E2E 183/183 passes, server /health returns 200. 18 分钟。

在代码冲突之前捕获意图冲突。一个构建在 Git 之上的开源 coding agents 协调协议。

测试发现：All 98 tests pass, both binaries (foremerge, fmg) report version 0.5.0, the query benchmark runs successfully, and the doctor command reports a healthy install with an uninitialized database (expected pre-init state). 3 分钟。

OpenCode 目标插件，用于 Codex 风格的目标模式、/goal 斜杠命令、持久目标以及 AI 编码代理焦点。

测试发现：The repo installs cleanly from source: all dependencies resolve, TypeScript typechecks without errors, ESLint reports no violations, the full test suite (296 tests) passes, the server bundles to dist/server.js (206.6 KB), and npm pack --dry-run confirms the correct publish artifact (6 files, 284.4 KB unpacked). 2 分钟。

用于浏览 models.dev、benchmarks、coding agents 以及 AI 提供商状态的 TUI 和 CLI。

测试发现：The modelsdev Rust project compiles, passes all 518 tests, and its CLI (help, version, list, agents status, status list, completions) produces correct output. 3 分钟。

Percho：为 Pi coding agent 打造的极简桌面 GUI，与 Pi CLI 使用相同的引擎，提供简洁的视觉界面。支持多会话聊天、可视化工具审批和自定义主题。

测试发现：All 538 tests pass across 49 test files in both packages/backend and packages/desktop; typecheck passes in all 3 workspaces; the production Electron build succeeds; the app launches and reports app ready v43.3.0. 8 分钟。

开源、Git 原生的 AI 编码代理记忆系统：为 Claude Code、Codex、Cursor、Gemini CLI 及兼容工具提供确定性本地回忆。

测试发现：All ownmem tests (structure, public, and tokenizer) pass; benchmark achieves 100% Recall@1 with deterministic local retrieval; CLI help and init --preview both respond correctly. 2 分钟。

Claude Code、Codex 和 multi-agent 工作流的跨运行时技能。

测试发现：109 installable skills validated with 0 errors, install.sh installs 110 skills to ~/.agents/skills for Codex, and the full 532-test suite passes. 13 分钟。

面向团队与朋友的多 Agent 协作

测试发现：After upgrading to Node v22.14.0 and passing --experimental-require-module --experimental-websocket, all 7 testable packages typecheck and test cleanly (5,284 passing, 2 pre-existing failures from missing bun and a better-sqlite3/kysely adapter mismatch). 51 分钟。

面向 coding agents 的开源 tmux GUI/TUI

测试发现：tmux-ide in this repo builds and runs real tmux 3.7c sessions with send/ls/adopt/team, a 200-serving command-center REST+SSE API, a working control-socket send verb, and the full managed dev-instance lifecycle; 6618 daemon unit tests, 551 contract tests, 273 daemon-client tests, 1103 desktop-renderer tests, 70... 37 分钟。

使用 Rust 编写的高性能 coding agent harness，专为大规模 multi-agent 开发工作流而构建。

测试发现：jcode v0.81.7-dev is built from source, installed via symlinks in ~/.jcode/builds/current and ~/.local/bin, and works correctly with OpenRouter provider. jcode run and jcode repl both complete successfully with real AI responses. 20 分钟。

将编码任务委托给一个独立的 coding agent CLI，审查 diff，自己提交 commit：每个实现者一个。

测试发现：The delegate-skills package installs all 18 skills, the test suite passes all checks against fake CLI shims, and the POSIX shim no longer depends on the external dirname command. 30 分钟。

面向 Claude Code 的 Agentic 开发工具，采用 SPEC 驱动的计划/运行/同步，TRUST 5 质量门，模型+工作量路由，以及 Claude×GLM 多 LLM 成本控制。单个 Go 二进制文件，支持 16 种语言，零依赖。

测试发现：moai-adk v3.1.3 builds and runs. Binary at ./bin/moai responds to 'moai version' with v3.1.3. Test suite runs with 3 pre-existing failures (2 test-isolation, 1 codex-binary not available). templ code generation works. All ast-grep rule tests pass with version 0.40.5. 57 分钟。

Token 高效的 Claude Code Python 完整重建。310K LoC Python 的 AI Coding Agent。

测试发现：Python virtual environment installs all dependencies cleanly, the 'clawcodex' CLI starts and reports version 1.6.0, and the full test suite of 10,435 tests passes with 0 failures. 51 分钟。

AI 编码助手缺失的 linter 和 lsp。验证 CLAUDE.md、AGENTS.md、SKILL.md、hooks、MCP。包含适用于所有主流 IDE 的插件，并支持自动修复。

测试发现：agnix 0.52.2 installed and working from both npm (user dir) and cargo source build; 4700+ tests passing across all 6 workspace crates; binary validates agent configs and produces text, JSON, SARIF, and GitHub annotations output formats with i18n support. 20 分钟。

用于构建 coding agent 的 Rust SDK。工具、流式处理、图、sub-agent 编排、MCP，作为可组合函数。

测试发现：The 19-crate Rust workspace builds cleanly, all 612 tests pass, and the abstract-cli binary responds to --help and --version commands. 6 分钟。

面向 AI 编码 CLI 的多智能体编排：Claude Code、Kiro、Codex 等，在隔离的 tmux 会话中协调。

测试发现：CAO server starts, responds 200 on /health and /docs, and passes all unit tests (2986+ passed) except for 4 pre-existing test-isolation wiki_lint failures and Node version-gated web UI build. 43 分钟。

LeanCTX 是 AI 系统的上下文网关。选择任务相关的上下文，应用支持的控件，连接本地知识并检查记录的证据。开源引擎：CLI、MCP、钩子和支持的代理路径。

测试发现：lean-ctx 3.10.1 release binary is built, installed at ~/.cargo/bin/lean-ctx, and fully functional with all CLI commands working. 65 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。

用于 Claude Code 的代码搜索 MCP。将整个代码库作为任何 coding agent 的上下文。

测试发现：Install, build, typecheck, lint config, and all 35 tests pass across the monorepo. The core indexing pipeline runs end-to-end with mock providers. The MCP server starts but requires real OpenAI API key and Milvus credentials for production use. 15 分钟。

一个用于 Grok API 的开源 coding agent。

测试发现：All 48 test files pass (248 tests), typecheck passes via tsc --noEmit, and CLI help displays correctly via bun run dist/index.js --help. 9 分钟。

通过询问代码的功能来查找代码。一个面向 coding agents 的 CLI，使用 Jev 来发现相关文件和源代码上下文。

测试发现：Repository builds, types check, lints clean. All 4 test suites pass (70 pass, 0 fail). End-to-end CLI search works against a mock provider: auth saves credentials, doctor verifies access, search returns file locations and source excerpts. 6 分钟。

AI 的形式化验证：一个 Lean 4 + Dafny 证明检查器，可捕获虚假的 AI 证明（奖励黑客行为），并实现编码 agent 的验证递归自我改进（RSI）。适用于数学、物理和软件工程的 AI。

测试发现：160 hermetic Python tests pass (170 skipped for missing Lean/Dafny toolchains); pip install completes; all modules import; 4 previously-failing snapshot-integrity tests fixed and passing. 81 分钟。

Claude Code 的自动化 TDD 强制执行

测试发现：TDD Guard installs, builds, and runs. All unit tests (770) and linter tests (24) pass. Core validation logic works with mock AI client. Jest, Vitest, Go, and pytest reporter integrations work. Validator integration and 6 language-specific reporter tests need real API credentials or additional language runtimes not... 53 分钟。

用你的提示词在一秒内创建 agent 应用。创建 LLM Agent 所需的一切：工具、提示词、框架和模型，都集中在一个地方。

测试发现：Install built a working venv with all dependencies. All 4 tests pass against a mock OpenAI server (2 LLM: OpenAIModel, OpenAIChatModel; 2 RAG: add_files, add_text). Streamlit app launches on port 8592 and returns HTTP 200. The demogpt CLI entry point uses subprocess('streamlit') which resolves only from the venv bin... 55 分钟。

Ovis（原 herdr web UI）：免费的 herdr 开源浏览器和手机客户端。为 Claude Code、Codex 及所有 agent 面板提供聊天和实时终端，支持通过 SSH 连接远程 PC，并支持 Web 推送提醒。

测试发现：bun install, typecheck, vite build, and 904/906 unit tests pass; server starts on port 7317 and serves frontend pages with HTTP 200; health endpoint returns 502 because the herdr backend is not available in this environment. 3 分钟。

面向人与 AI agents 的开源协作系统。共享上下文、协调任务，并与你的团队、云端 agents、Codex、Claude Code、Grok、Grok Bot、OpenCode、Antigravity 和 Cursor 并行工作。

测试发现：The repository installs, typechecks, lints, and builds cleanly; the full vitest suite passes (1502 passed) against a user-space PostgreSQL 17.10; the credential-free paired-relay contract smoke passes (113 tests); and the built CLI answers --version 0.11.0 and --help. 27 分钟。

10x - 借助多步骤 Superpowers，AI 编码速度提升最高 20 倍。开源 agent，支持智能模型路由、BYOK，完全自托管。

测试发现：Build of core, shared, and CLI packages succeeded. All 234 tests pass (184 core, 50 shared). CLI binary at apps/cli/dist/index.js answers --version (0.1.0) and --help. Web build failed due to out-of-memory (SIGKILL) during Next.js static page generation. 9 分钟。

cloudroom 是开源平台，可高效运行 1,000 多个 agent

测试发现：The Rust core (cloudroom v0.1.84) builds from source on Ubuntu 24.04, passes all 49 tests (24 unit + 25 integration), passes clippy with no warnings, is correctly formatted, and the compiled binary reports its version. The desktop GUI (gui/) could not be built because it requires Node 22.19+ and pnpm 9.15.9, but only... 29 分钟。

600 行 TypeScript 写成的超级迷你版 pi，让你轻松从 0 写出属于你的 pi-agent

测试发现：npm install, tsc build, and all 25 vitest tests pass across 5 test files; the LLM stream function and agent loop work end-to-end against a mock OpenAI-compatible API. 4 分钟。

天枢harness是一个的终端编程智能体，它要回答的核心问题是：模型何以稳定地交付，目标不漂移、完成有证据、验证有闭环，不说'应该修好了'。为此它在模型与真实世界之间建立一层认知执行环境（CVM），把目标、状态、证据、资源、权限与终止条件从对话历史中外部化，由运行时持续管理。同时对deepseek v4极度优化。

测试发现：The repo installs and builds on Node 24, its 19005-test suite passes with 0 failures, and the CLI runs end-to-end against a local mock OpenAI-compatible provider returning a real agent response. 19 分钟。

Autoprompt 是一种编码 agent 技能，可将 agentic 编码任务的失败率降低 45%。

测试发现：autoprompt-skill v2.0.0 CLI installs globally, all 256 core/provider/benchmark tests pass cleanly on Node.js 20.20.2 with Python 3.12 + PyYAML 6.0.3 + jsonschema 4.26.0. 31 分钟。

帮助 coding agents 在使用公共或私有库时从不犯错，同时不浪费上下文窗口。

测试发现：After npm install, npm run build, and npm run check all succeeded, the app builds to an executable dist/index.cjs; the HTTP server answered 200 'pong' on /ping and a full MCP session (initialize, tools/list, tools/call search/read, prompts) returned 200 with correct results against a mock Ref API, while the stdio... 3 分钟。

一个轻量级终端编码助手，具有类似 Claude Code 的工作流、工具循环和 TUI 架构，专为学习和实验而构建。多语言支持：现提供 TypeScript、Python 和 Rust 实现。

测试发现：MiniCode installs, compiles, passes all 270 tests, and launches in mock mode successfully. 5 分钟。

由 DeepSeek 构建，为 DeepSeek 打造：一个 Swift 原生 macOS coding agent

测试发现：DeepTide Rust workspace builds, CLI runs in local-echo mode and against a mock OpenAI endpoint, all 1314 tests pass, GUI launches on virtual display, npm package entrypoint works under Bun. 10 分钟。

CUGA 是一个面向企业的开源通用型 agent harness，支持在 Web 和 API 上执行复杂任务、OpenAPI/MCP 集成、可组合架构、推理模式以及策略感知功能。

测试发现：CUGA 0.4.1 installs successfully with uv, imports cleanly, and 2358 of 2362 unit tests pass with a dummy OPENAI_API_KEY; 3 test_sdk_citations failures are ordering-dependent and pass when run in isolation. 29 分钟。

DeepSeek coding agent 的 harness：持久化 Shell、Ultra 子代理、自动批准、Chrome MCP 和会话遥测

测试发现：Install and setup complete, 570/586 unit tests pass, dscode exec mode works end-to-end with mocked LLM, TUI app needs real TTY: this is a headless-container limitation, the binary itself is healthy. 13 分钟。

coding agent 的实时消息

测试发现：The project installs and builds cleanly on Node 22.17.0 with pnpm 10.33.0, its CI commands all pass, and its CLI and MCP server ran through a complete two-agent claim, prompt, reply, transfer, and review workflow in a scratch workspace. 4 分钟。

Whale：极快的、终端优先的 AI coding agent，面向 DeepSeek。约 98% 的提示缓存命中率，1M 上下文，MCP 工具，动态工作流。

测试发现：Whale binary built (40 MB) at bin/whale, all 53 Go test packages pass, help/version/doctor commands respond correctly; only missing a DeepSeek API key. 3 分钟。

一个自托管的 OpenCode 命令中心，将会话、git、终端和日程集成在一个 Web 应用中，可通过手机或桌面访问。

测试发现：Dependencies installed, backend builds and serves HTTP on :5003 returning 200, test suite passes 3081 of 3148 tests (145 of 161 files) with 13 failures caused by missing external tools (unzip, opencode binary, msb) and API-key-dependent services. 33 分钟。

跨平台微型虚拟机沙箱，适用于智能体工作负载

测试发现：Dependencies installed in a venv and the release bundle downloaded/verified; the Python tooling test suite passes (454 tests), OpenVMM runs its 4562-test unit suite with only the container's NUMA-seccomp-blocked mbind tests failing, the openvmm binary answers version/help, and the kernel plus all three initramfs... 78 分钟。

HERO = Hashing · Edge cases · Rubrics · Overbuild，这是 coding agents 过度防御的四种形态。一个粘贴即用的契约来阻止它们。适用于 Claude Code, Codex, Antigravity, Cursor, Copilot, Windsurf...

测试发现：Python virtual environment created with websockets 15.0.1; the Goal Pulse CLI parses arguments and all 18 behavioral tests pass, including real Unix WebSocket transport tests without a Codex daemon. 1 分钟。

Pi，一个 coding agent，提前编译为单个原生可执行文件。启动速度快 10 倍，CPU 使用减少 7 倍，内存使用减少 4.5 倍，相比基于 Node 的 Pi。支持 Linux、macOS、Windows、ARM64 和 Android（Termux）。

测试发现：Pi-Bolt 0.9.8 binary installed and verified (version, help, doctor, model listing all functional). Source packages build cleanly after one type fix. 95 script tests pass, 525+ coding-agent tests pass, and the binary is ready for use with real provider credentials. 29 分钟。

2-in-1 AI 开发者和项目经理。AI agent 在 Todoist 中规划整个项目，并逐任务编写代码。

测试发现：All unit tests pass (14/14), all 28 source modules import cleanly, main entry points (manager.py, single_task_coder.py) import without error, and Playwright Chromium browser is installed. 4 分钟。

面向 coding agent 的职位公告板 CLI。搜索并评估全球数百万条白领行业职位信息，每小时刷新一次。

测试发现：Pinloop CLI v0.8.2 builds from source with tsc, installs via npm, and runs 15 distinct commands (search, count, fetch, judge, pull, login, logout, account, viewed, guide, welcome, filter, billing, tab, message) against a mock API server returning realistic responses. 6 分钟。

可组合的多 agent shell

测试发现：Python test suite passes (72/72), Rust test suite passes (8/8), Rust release binaries compile and respond to version and help, Python backend serves health endpoint, npcsh shell initializes and connects to the backend after a one-line migration SQL fix. 17 分钟。

用 Golang 编写的 pi agent

测试发现：pigo builds from source with Go 1.27.2; full test suite passes (1524/1524); the binary runs, and headless text, stream-json, session resume, the --no-tui REPL, and the full-screen TUI all executed a real agent loop against a schema-correct local mock Chat Completions endpoint, executing the bash tool and printing the... 21 分钟。

面向企业环境限制下的个人 AI 助手，基于您已有的 coding agent、工具、会话和权限构建。

测试发现：Python 3.12 venv installed the tenx-productivity editable package, the 16-test suite passes, and the slack-polling trigger host and stand-up prep CLI run end to end against a local schema-correct Slack mock, so the same binaries are ready for a real Slack token/cookie and a real coding-agent CLI once those credentials... 13 分钟。

🧬 The Huxley-Gödel Machine

测试发现：Python venv install is complete, all 22 repository tests pass, all main modules import, the real SWE-bench dataset and TestSpec construction were verified, and hgm.py boots through configuration/source-copy to its Docker-dependent self-improvement step, where execution stops because this container has no Docker daemon... 31 分钟。

Metis 是一个 coding agent，可将 AI/LLM 编码性能提升 50% QQ群: 801193315

测试发现：npm ci installs cleanly, npm run build produces working dist/, CLI prints version 1.4.0 and full help, 268 of 275 test files pass in non-root headless Electron environment. 34 分钟。

Java 8+ agentic SDK：统一的 LLM 访问（OpenAI/Anthropic/DashScope/Doubao/DeepSeek...），统一的 AI 服务（图像/视频/音频...），Tool Calling、MCP、RAG、SKILLS、Agent Runtime、A2A、agent...

测试发现：All 10 modules build, install, and pass their test suites (1624+ total tests, 0 failures, 10 skipped). The CLI JAR starts and responds to --help and extension list. The docs-site Docusaurus build requires Node.js >=20 (has 18.19.1) which is a preexisting constraint. 19 分钟。

所有我的 YouTube 教程。这里的一切都是免费的，但如果你觉得这些有帮助，请在 YouTube 上关注我。

测试发现：The full just check suite exits 0 with PATH containing casey/just, terraform, go, and uv binaries. 7 分钟。

Agent技能使AI编程助手编写生产级机器人软件。ROS1、ROS2、设计模式、SOLID原则和测试，适用于Claude Code、Cursor、Copilot以及任何兼容SKILL.md的agent。

测试发现：The install.sh installer installs skills to targets correctly, the skills validator reports 0 errors, and the with-skills demo_recorder eval package imports without ROS2 with its pytest suite running 16 passed and 12 ROS2-dependent skips, with a mock-driven node lifecycle saving a valid episode JSON. 10 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。

一个用于记录和分析 Claude Code 和 Cursor AI 驱动编码会话的 CLI 工具。

测试发现：The project installs, builds with tsup (ESM), passes all 839 tests, and the CLI binary prints its help text when launched with --help. 24 分钟。

轻量级 agent 编排开发工具，存在于你的 repo 中

测试发现：CLI builds and all core commands work end-to-end for spec and issue management. Test runner (vitest) crashes with bus error on Node 18. Server fails to start due to ESM module resolution issues. Frontend has pre-existing TypeScript errors. Project requires Node >=20 for full compatibility. 36 分钟。

在你的云端运行的开源自主工程 harness。每个任务一个沙箱，内部没有真实机密。

测试发现：All three component test suites (CLI 293/293, harness 1101/1101, helm chart 126/126) pass. Control-plane 1597/1608 pass; the 8 failures all require PostgreSQL 18 which is not available in this container. The CLI binary responds to its help command listing 10 commands. 8 分钟。

使用 cheap-ag 将你的使用量提升 50x。Opus/Astra 计划中，一个极其便宜的模型编写每一行代码，测试决定发布什么。你的 $20 计划不再受到限制。一个 MCP 服务器，用于 Claude Code 和 Codex。

测试发现：npm install, TypeScript build, and all 392 tests pass on Linux x86_64 with Node v22.23.4, using a compiled lsof replacement and no API keys. 7 分钟。

UmaDev：一个像真实开发团队一样工作的 coding agent，指挥你已经在使用的 Claude Code / Codex / OpenCode。

测试发现：The Rust project builds, all 1466 unit/integration tests pass, the umadev binary runs and responds to --help, --version, doctor, init, and ci subcommands from a source build with no external API keys. 11 分钟。

本地优先的 MCP 插件，用于由 AI coding agents 进行持续软件质量审查，由 Jev 驱动。

测试发现：The repo installs and builds with Node 20, its 13-test suite passes, the built dist/server.js is registered with the Codex CLI, and it serves an MCP jev_review call end-to-end against a local schema-valid Jev mock. 8 分钟。

通过语音与 Claude Code 和 Codex 对话，从你的手机，在任何你的机器上。用于 pitupitu.chat 的开源 box agent；你的 agent 在 herdr 中运行。

测试发现：The box agent builds with all 91 tests passing, install.sh installed pitupitu 0.8.0 and herdr 0.9.5, and pitupitu connect dialed out with a valid protocol-5 hello and held the link, with the box starting a real herdr server. 8 分钟。

Claude Code、Codex CLI 和其他 AI coding agents 的共享标准和可复用技能。

测试发现：All 77 tests pass across 6 test files: agy_task_runner (25), grok_acp_task_runner (20), telegram_corpus (18), package_skill (7), update_readme_skills (6), and agent_runtime_sync (1). All scripts import cleanly. 2 分钟。

开源桌面应用，用于本地 coding agent。查看每个工具调用、diff 和终端命令。自带模型；会话留在你的机器上。

测试发现：The openagentd 2.26.0 backend server and web UI build cleanly, the suite passes 4728/4728 (12 skipped) plus 3168 web tests, the server answers 200 on /api/health/ready and /api/agents and /api/agents/registry, and a real chat turn completed against a local OpenAI-compatible mock, saving the assistant replies into the... 20 分钟。

一个用 C 语言编写的极简终端原生 coding agent。

测试发现：The binary builds cleanly at build/hax, reports version 6753f30, runs one-shot and JSON-stream prompts through the mock provider end-to-end including a real tool call, and passes all 121 unit and e2e tests plus make lint. 7 分钟。

TalkCody - 代码廉价，请展示你的谈话。免费开源 AI Coding Agent。

测试发现：Bun/Node/Rust toolchain installed user-level; frontend builds and its 2845 tests plus 99 API tests pass against a local Turso mock; the API server answers 200 on /health and /api/remote-agents/configs; only the Tauri desktop binary could not be built because glib-2.0/webkit2gtk dev packages are missing and cannot be... 24 分钟。

AI 编码工作站：Claude Code + Web UI + 8 个 AI CLI + 无头浏览器 + 50 多个工具

测试发现：All 285 Node.js unit tests, 5 Python notification tests, and the CLI persistence shell test pass. Repository structure is validated end to end with product facts contracts, security policy, documentation, and patch scripts all verified. 2 分钟。

面向 AGI、抗幻觉的 AI 编码运行时，基于证据、工具、记忆、agent 和验证。

测试发现：Install and build succeed CLI reports version 0 1 55 All commands render in help 6 of 13 failing test files fixed all remaining failures are pre-existing and confirmed in the original code. 74 分钟。

基于规范、智能体驱动的工作流框架，专为 AI 编码 agent 设计。将请求转化为可验证的目标循环：计划、执行、验证，并在你的仓库中保留持久的规范和证据。兼容 Claude Code、Codex、Gemini、OpenCode 以及普通 CLI。

测试发现：npm install completed without errors (Node 18 engine warnings only); ospec CLI v2.1.0 responds to --version, --help, init, status, change, session commands; release smoke test and content scan both pass; index rebuild tool works; global install via --prefix succeeds. 5 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。

为 AI agent 提供沙盒化代码执行，支持本地或云端。大规模并行，易于扩展。为 SWE-agent 等提供支持。

测试发现：Package installs cleanly, 74 tests pass, server binary starts and responds correctly on all core endpoints (is_alive, create_session, execute), CLI --version prints 1.4.0. 23 分钟。

Sno Station Skills：31 项经过测试的技能，可将 Claude Code 和 Codex 变成一个协作团队。跨供应商同行评审、当 agent 达到配额时自动交接、一页晨间简报、带有证据的工作单，以及从你自己的会话中学习的夜间循环。开源，通过一条 sno setup 命令安装。

测试发现：sno 1.1.13 is installed at /home/runner/.local/bin/sno with Sno Station setup into Codex; 19 of 22 unit selftests pass on this container, catch-report passes all 24 under a local Anthropic mock, and medic/rem-reflect selftests fail only on missing system tmux/strace which need root to install. 69 分钟。

用于 token 最大化的 agent 流程系统

测试发现：The hmz package installs cleanly, ruff lint passes, the CLI binary prints valid usage output, and 2600+ unit tests + 145+ integration tests pass against mocked backends; 11 unit test failures are pre-existing environment constraints (Landlock, argv0 path) unrelated to install. 25 分钟。

一个针对 Apple silicon 的本地推理引擎，围绕模型构建。

测试发现：Python server components (installer, launcher, model catalog, HTTP server, API protocol, images, documents, chat templates, constraints, tools, tokenization, crash tracing) pass ~1031 tests on Linux. Native Metal inference engine requires macOS 26.4+ Apple Silicon and Xcode. 34 分钟。

Agentic SDLC 基础设施。定义护栏。收集信号。持续执行。你的 AI coding agents 和 CI/CD 管道所做操作的签名记录，对照策略即代码进行检查，并在 CI/CD 中闭环。AI 会话、SBOM、VEX、SARIF 和 SLSA 出处，基于 Sigstore 和 in-toto...

测试发现：Chainloop CLI and both server components (controlplane, artifact-cas) compile and install. 71 of 76 test suites pass. The 5 failing suites require Docker/testcontainers which is unavailable. The installed CLI binary responds to --help and version commands. 16 分钟。

开源 AI 代码审查工具，专为编码 agent 设计。在 Claude Code、Codex、Cursor 和 Cline 中，独立的审查者会在你推送之前检查 agent 编写的每一次更改。无需自己的账户或密钥。

测试发现：With Node 22.20.0 and pnpm 9.15.4 under $HOME/opt: install and build are clean, the CLI launches and reports 0.8.1, init installs working agent files, scan runs 8 real scanners and reports the demo's 22 planted findings, unit and e2e suites pass fully except two documented environment-version cases (git 2.43 vs 2.44... 13 分钟。

自主软件工程 AI agents 舰队，用于在 AgentField 上生成生产级 PRs：计划、编码、测试和交付。

测试发现：The swe-af package is installed in the venv, all 1317 tests pass, and python -m swe_af starts a uvicorn server on port 8001 (reports degraded mode because no AgentField control plane is running on localhost:9999). 14 分钟。

Poirot 是一个深度研究 agent 内核，专为关注 agent 架构设计的人打造。

测试发现：poirot installs via pip into a venv, all 2716 unit/integration tests pass, and the CLI binary reports help for run and cli subcommands. 30 分钟。

Agent OS：agent 会自行变得更聪明。我们只需守住底线：面试门槛、分阶段评估、预算化进化循环。MCP server，14 个运行时：Claude Code、Codex CLI、Gemini CLI、OpenCode、Copilot、Kiro 等。

测试发现：All unit tests pass (0 failures in ~21000 tests across 34 unit directories), ruff lint and format clean, CLI boots with correct version, MCP server help displays correctly, 2 Windows-specific bugs were caught and fixed. 73 分钟。

面向开源模型（如 Kimi K3 和 GLM 5.3）的 coding agent

测试发现：The codex-rs workspace builds successfully, the binary launches, and the core protocols/config/otel/api tests all pass. CLI tests pass at 99.4% with only 2 pre-existing failures unrelated to the installation: features list doesn't fetch cloud config, and sandbox enforcement needs bubblewrap. 71 分钟。

AI agents 引导你从想法到可运行的代码，每一步都由你掌控。

测试发现：nwave-ai v3.22.2 installed in a venv, deploys to ~/.claude/ for Codex CLI, all 10 doctor checks pass, 34 agents and 151 skills install successfully, and all key Python modules (nwave_ai.cli, des.domain, nwave_ai.sync, nwave_ai.outcomes) import without errors. 7 分钟。

鸿蒙（HarmonyOS NEXT）开发知识包，一份源文件自动产出 Claude Code / Cursor / Copilot / Codex / Gemini CLI / Windsurf / ChatGPT 等 11+ AI 工具的配置文件，让 AI 像熟读华为文档的工程师一样帮你写 ArkTS / ArkUI。

测试发现：Build pipeline produces 29 dist files for 11+ AI tool formats from harmonyos-development/SKILL.md (4461 lines, 65 sections). All validation and frontmatter checks pass. Claude Code skill symlinked at ~/.claude/skills/harmonyos-development. 10 分钟。

轻量级、无容器的沙箱，用于运行具有网络和文件系统限制的命令

测试发现：无法验证；log 显示了它停在哪里。 10 分钟。

在你的终端中进行 AI 结对编程：一个静态二进制文件，亚毫秒级启动，支持任意模型。

测试发现：Prebuilt agentty 0.9.26 linux x86_64 release binary is installed at /home/runner/.local/bin/agentty and responds to all query subcommands (version, status, config doctor, skills, help, diagnostics). Source build fails at the jaal submodule due to insufficient compiler support for C++26 structured binding packs. 30 分钟。

面向智能体系统的模型路由器。在 50ms 内将每个提示路由到正确的模型。仅需更改端点即可将成本降低 40-70%。

测试发现：无法验证；log 显示了它停在哪里。 14 分钟。

使用 Rust 编写的开源终端 agent。可使用任何提供商或模型。始终欢迎 Issues 和 PR。

测试发现：无法验证；log 显示了它停在哪里。 19 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。

向 agent 展示问题：标记、录制、发送

测试发现：无法验证；log 显示了它停在哪里。 47 分钟。

终端中的 AI 结对编程：单个静态二进制文件，亚毫秒启动，支持任何模型

测试发现：无法验证；log 显示了它停在哪里。 7 分钟。

对抗性 AI 漏洞猎手，具有自动修复能力，适用于 Claude Code、Cursor、Codex CLI、GitHub Copilot CLI、Kiro CLI、Opencode、Pi Coding Agent 等。多 agent 管道可发现安全漏洞、逻辑错误和运行时错误，然后在安全分支上自主修复。

测试发现：无法验证；log 显示了它停在哪里。

面向高级 AI 编码工作流的开源软件工厂基础设施

测试发现：无法验证；log 显示了它停在哪里。

Orca Agent, a DeepSeek-native terminal coding agent

测试发现：无法验证；log 显示了它停在哪里。

具备组织治理的规格驱动开发：规格告诉 AI agents 要构建什么，章程规范其构建方式；Git 原生任务、可执行的工作流，以及跨 AI coding agents 和 harnesses 的一致工程标准。

测试发现：无法验证；log 显示了它停在哪里。 87 分钟。