Coding Agent Integration

Checksum integrates directly with AI coding agents like Claude Code, Cursor, and other tools via slash commands. Instead of using the cloud-based web app, developers can run /checksum commands in their IDE to detect, generate, and heal tests locally.

This is a complementary workflow — use the web app for managing your test suite at scale, and coding agent integration for quick, targeted test work during development.

---

Supported Tools

• Claude Code
• Cursor
• Augment Code
• IntelliJ (via Junie)
• Anti Gravity
• Rovo Dev
• Custom tools (any IDE that supports slash commands)

---

Installation

Step 1: Set Up Your Test Repository

First, make sure your Checksum test repo is ready:

# Clone your test repo
npm install
npx playwright install --with-deps
npx checksumai dotenv --download --api-key=<YOUR_API_KEY>

# Verify setup
npx checksumai test -g "example"

Step 2: Install Prerequisites

Make sure you have ripgrep installed:

brew install ripgrep

Step 3: Install the Agent

Run the installation script:

curl -fsSL "https://storage.googleapis.com/checksum-agent-public/install.sh?$(date +%s)" | bash

Follow the on-screen instructions. The installer will ask which coding tool you use and configure the slash commands automatically.

Claude Code and Cursor are detected and configured automatically. For other tools, you’ll need to specify the slash commands directory (see below).

Custom Tool Setup

If your tool isn’t listed, choose “custom” and enter the path to your tool’s slash commands directory:

Tool	User Scope	Project Scope
Augment Code	~/.augment/commands/	.augment/commands/
IntelliJ (Junie)	~/.junie/commands	.junie/commands
Anti Gravity	~/.gemini/antigravity/global_workflows	.gemini/antigravity/global_workflows
Rovo Dev	~/.rovodev	.rovodev

Some tools require their CLI to be installed separately (e.g., Junie for IntelliJ, Auggie for Augment Code).

Rovo Dev Additional Setup

For Rovo Dev, you also need to create a prompts.yml file in the .rovodev folder:

prompts:
  - name: Checksum_Detection_Agent
    description: Checksum Detection Agent
    content_file: checksum_detect_tests.md

  - name: Checksum_Generation_Agent
    description: Checksum Generation Agent
    content_file: checksum_generate_tests.md

  - name: Checksum_Healing_Agent
    description: Checksum Healing Agent
    content_file: checksum_heal_tests.md

Step 4: Verify

Open your coding tool and type /checksum. You should see three commands:

• /checksum_detect_tests
• /checksum_generate_tests
• /checksum_heal_tests

---

The Three Commands

/checksum_detect_tests — Detect Test Cases

Analyzes your codebase to identify the most important test cases to create.

Include in your prompt:

• Path to your code repository
• Path to your test repository
• Which features to focus on (and which to exclude)
• Number and type of tests (e.g., “2 marquee tests, 8 regular tests”)
• Any specific test data or files needed

Example:

/checksum_detect_tests the code repo is ./source-code the test repository is ./checksum-tests.
Find the most important features other than Settings.
Detect 10 test cases: 2 marquee tests and 8 regular tests.

Marquee tests are longer end-to-end full-flow tests. Regular tests are smaller, focused tests.

What happens:

1. The agent explores your repositories
2. It may ask clarifying questions — answer them
3. Once you see tool calls (bash commands), the agent runs autonomously
4. Output: .md test definition files that feed into generation

---

/checksum_generate_tests — Generate Tests

Takes the output of detection and generates Playwright test code.

Include in your prompt:

• Path to your code repository
• Path to your test repository
• Test plans from detection (file paths) or paste test cases directly
• Special login instructions (if not using the default)
• Any new credentials or test data

Example:

/checksum_generate_tests the code repo is ./source-code the test repository is ./checksum-tests.
Generate tests from the markdown files in ./checksum-tests/detected-tests/.
See existing tests in ./checksum-tests/tests/payment/ for reference.

Output: .checksum.md story files and .checksum.spec.ts Playwright test files.

---

/checksum_heal_tests — Fix Broken Tests

Updates and fixes existing tests after code changes or when tests start failing.

Include in your prompt:

• Path to your code and test repositories
• Which tests need healing
• Context for why healing is needed

Example (after code changes):

/checksum_heal_tests the code repo is ./source-code the test repository is ./checksum-tests.
Heal the payment tests in ./checksum-tests/tests/payment/.
Changes are committed to branch feature/new-checkout, compare against main.

Example (failing tests):

/checksum_heal_tests the code repo is ./source-code the test repository is ./checksum-tests.
The following tests are failing: checkout.spec.ts and cart.spec.ts.

---

Tips

• Allow exploration — Let the agent explore your repositories and answer its questions for better results
• Monitor initial tool calls — Watch until you see bash commands with tool calls, then the agent can run autonomously
• Reference existing tests — Point the agent to existing tests as examples of your coding patterns
• Provide context — The more context you give (PRD links, ticket descriptions, specific features), the better the output