NOASSERTION 许可；带执行记忆的自然语言 Web/移动端自改进测试框架，通过 MCP 工具与技能接入编码智能体；README 未明确提及 DeepSeek Harness，dsh 集成属推断。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
NOASSERTION
01它能帮你完成什么？
Author natural-language regression tests for web and mobile apps and run them from an agent-driven workflow
Self-healing test runs that recover from UI drift and flaky interactions, plus execution memory that improves future runs and catches regressions before shipping
QA engineers and software teams that want to write tests in plain language instead of brittle selectors
Wire environment setup, API calls, fixture seeding and teardown into test runs via sandboxed hooks
Node, Bun, Python or Bash hooks executed in isolated Docker containers, with structured outputs passed back into the active test run
Teams that need scripted environment or data preparation around UI tests
02如何接入 DeepSeek Harness？
前置条件
- Node.js project with npm — agent-qa is installed as a dev dependency via `npm install -D agent-qa`
- Docker installed before using hooks — Docker is required for the Node, Bun, Python and Bash hook containers
- Optional: `@vostride/agent-qa-subscription-auth` for Codex or Claude Code subscription auth
安装步骤
- 01
`npm install -D agent-qa`
$ npm install -D agent-qa
- 02
Optional (Codex or Claude Code subscription auth): `npm install -D @vostride/agent-qa-subscription-auth`
$ npm install -D @vostride/agent-qa-subscription-auth
- 03
`npx agent-qa init`
$ npx agent-qa init
- 04
`npx agent-qa install-browsers --chromium`
$ npx agent-qa install-browsers --chromium
- 05
Mobile projects: `npx agent-qa install-mobile-drivers --all`
$ npx agent-qa install-mobile-drivers --all
- 06
`npx agent-qa dashboard --open` — start the dashboard, complete auth, and run tests from the UI
$ npx agent-qa dashboard --open
验证接入成功
作者未说明
03DSH 适配与能力边界
Exposes MCP and skills for coding agents (per README); DeepSeek Harness/dsh is not mentioned in the README, so usage inside dsh is inferred via MCP
Natural-language tests for web and mobile
test definitions written in human language (actions and assertions)→agents execute tests from visible roles, labels, and screen state
launches installed browsers / mobile drivers to execute test runssends test context to the LLM endpoint you configureSelf-healing execution with memory
failing sub-actions (click, fill, select) and past-run observations→re-observed UI and alternate paths within the same run; execution memory added to future runs
persists execution memory built from product, suite, and test observations for future runsSandboxed test hooks
Node, Bun, Python or Bash hook scripts→hooks run in isolated Docker containers; structured outputs passed back into the active run
runs Docker containers for hooks (Docker required)Bring your own LLM + dashboard/CLI/MCP
any OpenAI-/Anthropic-compatible endpoint, Gemini, local or open-source model, or Codex / Claude Code subscription→tests executed with the model of your choice; dashboard and CLI for humans, MCP and skills for coding agents
04适合谁？何时不该用？
适合
- QA engineers and software teams that want to write tests in plain language instead of brittle selectors
- Teams that need scripted environment or data preparation around UI tests
不适合
- Hooks require Docker — agent-qa runs Node, Bun, Python and Bash hooks in isolated Docker containers, so Docker must be installed before using hooks.
- The README never mentions DeepSeek Harness or dsh; running agent-qa inside dsh is inferred from the stated MCP and skills support for coding agents, not documented by the project.
05兼容性、维护与安全提示
- Hooks require Docker — agent-qa runs Node, Bun, Python and Bash hooks in isolated Docker containers, so Docker must be installed before using hooks.
- No bundled model — you must bring your own LLM via OpenAI-/Anthropic-compatible endpoints, Gemini, local or open-source models, or Codex / Claude Code subscriptions, which means network access and your own credentials/subscription.
- The README never mentions DeepSeek Harness or dsh; running agent-qa inside dsh is inferred from the stated MCP and skills support for coding agents, not documented by the project.
NOASSERTION · actively maintained (last push 2026-08-03)
06常见问题
agent-qa 官方支持 DeepSeek Harness 吗？
README 没有提及 DeepSeek Harness 或 dsh。agent-qa 提供面向编码智能体的 MCP 与技能，并支持自带模型，因此可以作为基于 MCP 的测试工具在 dsh 旁使用，但项目本身并未记录这一集成。
如何安装 agent-qa？
运行 `npm install -D agent-qa`，然后执行 `npx agent-qa init` 并安装运行时支持，例如 `npx agent-qa install-browsers --chromium`（移动端项目：`install-mobile-drivers --all`）。用 `npx agent-qa dashboard --open` 启动仪表盘。若使用 Codex 或 Claude Code 订阅授权，还需安装 `@vostride/agent-qa-subscription-auth`。
必须安装 Docker 吗？
只有使用钩子时才需要——agent-qa 在隔离的 Docker 容器中运行 Node、Bun、Python 和 Bash 钩子，使用钩子前必须先安装 Docker。
可以用哪些模型跑测试？
支持 OpenAI 或 Anthropic 兼容接口、Gemini、本地或开源模型，以及 Codex、Claude Code 等订阅，任选其一。
测试和记忆存放在哪里？
测试、配置、钩子、记忆与套件逻辑都以版本控制的代码形式存放在你的仓库里，任何改动都可以 diff、评审、复用并跨团队共享。
07相关的 DSH 工作流
ruflo
作者 ruvnet
原版智能体元框架（meta-harness）：支持多智能体集群部署、自主工作流编排与对话式 AI 系统构建，内置自适应记忆、自学习智能与 RAG 检索，原生集成 Claude Code / Codex / Hermes 等大量工具。
reactive-resume
作者 amruthpillai
注重隐私的开源简历生成器，安全、可定制、可移植，永久免费，支持自托管。
everos
作者 evermind-ai
为所有 AI 智能体提供的便携统一记忆层，本地优先、Markdown 原生、数据归用户所有，跨应用与工作流自进化。
yao
作者 yaoapp
在桌面、移动端、浏览器或 API 上统一管理你的智能体与工作区，看板式任务追踪，支持自托管。
08数据与来源
**Built for humans and machines**: A polished dashboard and CLI for developers, plus MCP and skills for coding agents.
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。