Local installation packages

Put the book inside your coding agent.

Choose the agent you already use. The bundles contain the Testing AI references and book workflow locally. No separate API key is required to apply the book to a project.

Slash-command hosts/testing_ai assess the current diff and tests
OpenAI Codex$testing-ai assess the current diff and tests
No coding agent yet? Start here →
The Testing AI book connected inside an AI coding agent, recommending repeated trials and a chi-squared comparison between builds.
The Testing AI book works from inside the coding agent, applying its methods directly to the code, tests, results, and evidence already in context.

New to AI coding agents

Start with Claude Desktop. Then give it your project.

You do not need to be a developer. Install Claude Desktop, sign in, select the folder that contains the work you want assessed, and then add the Testing AI plugin.

Recommended for testers · Anthropic

Install Claude Desktop

Claude Desktop includes Claude Code. Sign in, open the Code tab, choose Local, and select your project folder. No terminal is required.

Get Claude Desktop
Step 1

Install Claude Desktop

Download the desktop app above and sign in. It can work with the files you explicitly select.

Step 2

Select one project folder

Open the Code tab, choose Local, then Select folder. Choose the directory where the project and its testing evidence live.

Step 3

Install Testing AI

Continue to the matching agent card below, install the plugin, start a new task, and run the first command.

What should be in the project folder?

Use the folder where the real work already lives. You do not need to make a duplicate. The more relevant evidence the agent can read, the more specific the book-guided assessment can be.

  • Source code, prompts, and configuration
  • Requirements, designs, and release notes
  • Tests, fixtures, and coverage reports
  • Eval datasets, rubrics, and judge prompts
  • Results, failures, logs, and traces
  • Metrics, screenshots, and bounded data samples

Keep secrets out of the folder context: do not include API keys, passwords, private customer data, or unrestricted production datasets. Redact or provide a safe sample when necessary.

Supported coding agents

Install once. Open your project. Run the book.

Each download is included beside this page, so the folder can be uploaded or served without nested paths.

Claude

Chat, Cowork, and Claude Code

Use Claude’s plugin upload screen. The same package also runs from Claude Code.

  1. Download testing-ai-claude.zip.
  2. In Claude, open Customize → Plugins.
  3. Choose Add plugin → Upload plugin, select the ZIP, then choose Install.
  4. Open your project folder and run /testing_ai.
Download for Claude
Claude Code terminal fallback
cd "$HOME/Downloads"
unzip -o testing-ai-claude.zip
cd "/path/to/your/project"
claude --plugin-dir "$HOME/Downloads/testing-ai"
OpenAI Codex

Codex plugin marketplace

Add the included local marketplace, install the plugin, and begin a task in your project.

  1. Download and extract testing-ai-codex.zip into a folder named testing-ai-codex.
  2. In ChatGPT, choose Codex → Plugins → Add marketplace.
  3. Choose Local folder and select the extracted folder.
  4. Install Testing AI, start a new task, and run $testing-ai.
Download for Codex
Codex terminal fallback
cd "$HOME/Downloads"
unzip -o testing-ai-codex.zip -d testing-ai-codex
codex plugin marketplace add "$HOME/Downloads/testing-ai-codex"
codex plugin add testing-ai@testing-ai-local
Cursor

Project skills

Copy the portable skills into the current project’s Cursor skill directory.

  1. Download testing-ai.plugin into your project folder.
  2. Extract it into a temporary testing-ai folder.
  3. Copy its skills into .cursor/skills.
  4. Restart Cursor, open the project, and ask it to apply Testing AI.
Download for Cursor
Copy-ready setup
unzip testing-ai.plugin -d testing-ai
mkdir -p .cursor/skills
cp -r testing-ai/skills/* .cursor/skills/
Antigravity

Gemini local plugin

Install the portable bundle in Antigravity’s local Gemini plugin directory.

  1. Download testing-ai-antigravity.plugin.
  2. Create the local plugin folder shown below.
  3. Extract the package into that folder.
  4. Restart Antigravity and ask it to apply Testing AI to the open project.
Download for Antigravity
Copy-ready setup
mkdir -p "$HOME/.gemini/config/plugins/testing-ai"
unzip testing-ai-antigravity.plugin -d "$HOME/.gemini/config/plugins/testing-ai"
Any MCP client

Standalone local server

Use the local stdio server when your agent supports MCP but does not support these plugin formats.

  1. Download and extract testing-ai-mcp.zip to a permanent folder.
  2. Open the agent’s Settings → MCP servers screen.
  3. Add a local stdio server with command node.
  4. Use the absolute path to testing-ai-mcp/bin/testing-ai-mcp.mjs as the only argument.
Download standalone MCP
Copy-ready MCP configuration
{
  "command": "node",
  "args": ["/absolute/path/to/testing-ai-mcp/bin/testing-ai-mcp.mjs"]
}
Command namespace: use /testing_ai in slash-command hosts and $testing-ai in Codex. This standalone plugin focuses entirely on applying the book.
Credentials: applying the bundled book references locally does not require a separate API key. An optional managed testing service can require credentials, but it is not needed for the book workflow.
Privacy: Testing AI records anonymous page views, install/download actions, one first plugin activation, and the allowlisted workflow name: assess, review, plan, apply, or catalog. It never sends prompts or project content. Set TESTING_AI_ANALYTICS=off before starting your coding agent to disable plugin analytics.