Documentationbreadcrumb arrow Grafana Cloudbreadcrumb arrow Testing and syntheticsbreadcrumb arrow Performance testingbreadcrumb arrow Author and run testsbreadcrumb arrow Script Authoring mode
Grafana Cloud

Use k6 Script Authoring mode

k6 Script Authoring is an AI-powered mode of the Grafana Assistant that helps you create production-ready k6 performance test scripts. Instead of writing code from scratch, you describe what you want to test and Assistant generates a complete script that follows k6 best practices—including proper URL grouping, checks, thresholds, and built-in tracing and profiling instrumentation.

k6 Script Authoring mode overview

Before you begin

k6 Script Authoring mode requires Grafana Assistant to be enabled in your Grafana Cloud instance. Contact your administrator if you do not see the Assistant icon in the sidebar.

Access k6 Script Authoring mode

Follow these steps to open the Grafana Assistant in k6 Script Authoring mode.

  1. On the main menu, click Testing & synthetics > Performance.
  2. Click the Assistant icon at the top of the sidebar to open the Grafana Assistant.
  3. In the mode selector at the bottom of the Assistant panel, choose k6 Script Authoring.

After the mode loads, the Assistant presents three workflows:

  • Write a new test: Describe the service or endpoints you want to test; Assistant generates a script.
  • Discover and generate: Provide a service name; Assistant queries your metrics, logs, and traces to find real endpoints and traffic patterns.
  • Work with an existing script: Paste, attach, or open a script you already have; Assistant analyzes, improves, or converts it.

If your first message is already a clear request, Assistant skips the menu and begins immediately.

Write a new test

You describe the load scenario in plain language; the Assistant returns a script you can review and run.

Try a prompt like this for a fictional API:

“Generate a load test for https://api.example.com. Test the GET /products and POST /orders endpoints with 20 virtual users for 5 minutes.”

The Assistant generates a complete k6 script based on your description, including:

  • HTTP requests to the specified endpoints
  • Realistic load configuration (VUs, duration, stages)
  • Response checks and thresholds
  • URL grouping to prevent metric cardinality issues
  • Tempo tracing and Pyroscope profiling instrumentation

You can also provide additional context to improve the generated script:

  • Authentication details: Describe the auth scheme (bearer token, API key, basic auth) and the Assistant adds the correct headers and setup code.
  • OpenAPI/Swagger spec: Paste the spec, attach it as a file, or provide a URL. The Assistant uses it to fill in accurate request bodies, parameters, and response checks, reducing the number of placeholder // TODO: comments.

Discover and generate

Use this workflow when you want Assistant to find real endpoints and traffic patterns from your Grafana observability data, or from your source code repository. This produces a more accurate script because the load profile is based on what your service actually handles in production.

Provide a service name

Start by prompting Assistant with the name of the service you want to test:

“Generate a k6 script for the checkout service”

Assistant queries your telemetry to generate the script.

Review and confirm endpoints

After discovery, Assistant lists endpoints by category, for example read operations, write operations, or internal and health probes. Each line usually includes RPS and p95 latency:

I found 9 endpoints for service "checkout":

Read operations
1. GET /api/v1/products/:id      210 RPS · p95 55ms
2. GET /api/v1/cart/:cartId       45 RPS · p95 120ms
3. GET /api/v1/orders/:orderId    12 RPS · p95 95ms

Write operations
4. POST /api/v1/orders             8 RPS · p95 380ms
5. PUT  /api/v1/cart/:cartId       6 RPS · p95 210ms
6. DELETE /api/v1/cart/items/:id   2 RPS · p95 180ms

Internal / health
7. GET /health                   520 RPS · p95 2ms
8. GET /metrics                   80 RPS · p95 5ms
9. GET /internal/cache/stats       1 RPS · p95 8ms

My suggested selection: 1, 2, 4, 5 (covers ~90% of observed traffic, read + write mix)
- Skipping health probes (#7, #8) — they inflate RPS and are not meaningful to stress-test
- Skipping #9 — internal endpoint, < 1 RPS

Does this look good, or would you like to add or remove any endpoints?

You can reply with endpoint numbers, plain language (for example, “only write operations” or “add the delete endpoint”), or just confirm the suggestion.

Enrich with an OpenAPI spec

After you confirm the endpoint selection, you can supply an OpenAPI or Swagger spec for the service when you have one. That input is used to fill in:

  • Path and query parameters
  • Request body schemas
  • Authentication schemes
  • Expected response codes and fields for check() assertions

You can provide the spec in any of these ways:

  • File attachment: Use the attachment button in the chat input to attach a JSON or YAML spec file.
  • Paste: Copy and paste the spec content directly into the chat.
  • Public URL: Provide the URL so the spec can be fetched from a public location.
  • Private or internal URL: Use your connected Infinity data source to fetch a spec from an internal network.

If you don’t have a spec, the Assistant proceeds with the observability data alone and marks fields that need your review with // TODO: comments.

Add a GitHub repository as context

If you provide a GitHub repository URL, Assistant uses the repository as its primary source of context instead of querying observability data first.

“Generate a k6 script for https://github.com/my-org/checkout-service"

The Assistant fetches relevant files from the repository—route handlers, API client code, authentication middleware, request/response schemas, and existing tests—and uses those to drive script generation. This workflow suits services that are not in production or that don’t send telemetry to your Grafana Cloud stack yet.

Work with an existing script

Use this workflow to analyze, improve, or convert an existing k6 script.

Analyze a script

Paste a script into the chat or attach one, then prompt for analysis, tuning, or conversion from another format.

  • High metric cardinality: Dynamic URLs used directly as metric labels can produce thousands of unique time series, which causes performance problems in Grafana Cloud. Assistant flags these and rewrites them using http.url template literals or tags: { name }.
  • Missing checks: HTTP requests with no response validation mean test failures go undetected. Assistant adds check() calls to validate status codes and key response fields.
  • Missing thresholds: Without thresholds, a test has no pass/fail criteria. Assistant adds thresholds to the options block based on the endpoints and response time data it finds.
  • Incorrect lifecycle usage: HTTP requests placed in the init context run once globally instead of per virtual user. The Assistant moves them to the correct VU code function.
  • Invalid test options: Certain option combinations that are valid in k6 OSS fail validation in Grafana Cloud k6 (for example, mixing duration with iterations, or setting startVUs: 0). The Assistant may identify these and suggest corrections.

Improve a script

After analysis, prompt Assistant to apply the fixes:

“Fix the issues you found and improve the load profile based on our production traffic”

The Assistant rewrites the script with all corrections applied and explains what changed.

Convert a script

You can convert existing test artifacts into k6 scripts:

  • Postman collection: Paste or attach your Postman collection JSON and Assistant converts requests, environment variables, pre-request scripts, and tests into the equivalent k6 code.
  • curl commands: Paste one or more curl commands and Assistant generates the corresponding k6 HTTP calls, including headers, authentication, and request bodies.