testmesh generate

The generate command uses AI to create TestMesh flow YAML from an OpenAPI spec, a running service, or a plain-English description.

Usage

# From an OpenAPI spec file
testmesh generate --spec ./openapi.yaml

# From a running service (auto-discovers endpoints)
testmesh generate --url http://localhost:5001

# From a description
testmesh generate --description "Test creating and retrieving a user"

generate requires an AI provider API key. Set ANTHROPIC_API_KEY or OPENAI_API_KEY in your environment before running this command.

export ANTHROPIC_API_KEY=sk-ant-...

Options

Flag	Description
`--spec <file>`	Path to an OpenAPI 3.x YAML or JSON spec
`--url <url>`	Base URL of a running service to introspect
`--description <text>`	Natural language description of what to test
`--output <file>`	Write the generated flow to a file (default: stdout)
`--model <model>`	AI model to use (default: provider's recommended model)

Examples

From an OpenAPI Spec

testmesh generate --spec ./openapi.yaml --output ./flows/api-tests.yaml

The generator reads the spec, identifies endpoints and their request/response schemas, and produces a flow that exercises each operation with appropriate assertions.

From a Running Service

testmesh generate --url http://localhost:5001 --output ./flows/user-service.yaml

TestMesh queries the service for available routes (via /openapi.json or similar) and generates a flow. Works best with services that expose an OpenAPI spec at a standard path.

From a Description

testmesh generate \
  --description "Test that a user can register, log in, and view their profile" \
  --output ./flows/auth-flow.yaml

The AI generates a plausible multi-step flow based on the description. Review the output and adjust endpoint URLs and assertions to match your actual API.

Refining Generated Output

Generated flows are a starting point. After generating:

Open the output file and verify the endpoint URLs match your service.
Adjust request bodies and headers as needed.
Tighten assertions to match the exact shape of your responses.
Run with testmesh validate then testmesh run to verify.

For conversational, iterative refinement, use testmesh chat instead. It lets you describe your requirements in natural language and refine the generated flow through back-and-forth conversation.