1. Docs
  2. AI
  3. MCP & AI assistants
  4. Example prompts

Example prompts and workflows

Ask in plain words, in your own language. Name the project, say what you want, and give the assistant a source to work from. These examples show what works and what the assistant does with each request.

4 min read Updated CustomersAdmins

In short

On this page you can

  • Copy prompts that get a project from empty to running pipelines.
  • Ask why a run went red and get the failed step and its cause.
  • Decide auto-heal records and handle defects from the chat.
  • Sync with Jira, read the test mailbox and query a test database.

How to use it

  1. Start with the project status prompt.
  2. Work through the stages in order, one module at a time.
  3. Approve each diff the assistant shows.
  4. Come back to the status prompt to see what is left.

First steps#

Start here
Which SmartestQA projects can I access?

Where does project Web Shop stand, and what should we do first?

The assistant calls smartest_context, then smartest_status, and lists the open work as work keys with counts. In Claude Desktop you can also start the Status prompt from the + menu.

Give it a source

The assistant writes better tests when it can read something real: your code (Claude Code or Cursor opened in the repository), the running app, an analysis document in SmartestQA, or what Discover already knows. Say which one: “Use the code in /web/src.”

Build a project, module by module#

StageAskThe assistant
Environment“Check the environment of Web Shop: language, start URL and where runs happen.”Proposes what is missing (level=environment).
Modules“Group the recorded pages of Web Shop into modules by business area.”Proposes modules and which pages go where.
Pages“List the pages without a purpose and draft one for each from the code in /web/src/pages.”Proposes page purposes and roles (level=page).
Requirements“Write the main requirements and sub-requirements of module Orders from the document Orders spec.”Reads the document in parts and proposes a requirement tree.
Scenarios“Plan scenarios for the sub-requirements of Checkout that have none, riskiest first.”Proposes a scenario plan per sub-requirement; after you approve, writes and dry-runs each one.
API“Build the API inventory of module Orders and add a refusal scenario for every write.”Proposes the inventory (level=api), then API and security scenarios.
Pipelines“Build the pipelines of module Orders.”Runs smartest_module_pipelines: functional, API and security pipelines, plus Smoke and E2E entries.
Schedule“Schedule the Orders functional pipeline on weekdays at 02:00, Berlin time.”Sets the cron once the pipeline has passed once.
Load tests“Propose load tests for the critical endpoints of Orders.”Proposes a load profile (level=perf).

The assistant follows this order on its own (see The order the assistant works in). You can also paste a command copied with With MCP from the Setup Roadmap.

Write and run one scenario#

Examples
Write and dry-run the planned scenario "Card is declined" on the server.

Run scenario 482 again on my agent; I fixed the test user.

The scenario "Export invoice" is BLOCKED. What are our options?

Dry runs on your agent need it online (see Install an agent). For a MOBILE project with Android and iOS, say which platform.

Read results#

Examples
Why did last night's Regression pipeline go red? Group the failures by cause.

Show the failed step, its locator and the screenshot of the latest run of "Login with SSO".

Which scenarios of module Orders flip between green and red?

What is running right now, and are the runners healthy?

The assistant uses smartest_runs. The Red runs prompt starts the same triage for the whole project.

Heal, defects and data#

Examples
Which healed locators are waiting for a decision? Show me each one before you approve it.

List the open defects by priority and offer to close the ones whose scenario ran green.

Open a defect for the failure in run 9031; it is a real product bug.

Which data keys have no value, and which scenarios use them?

Healed locators are approved one by one (preview first). New defects are proposed; closing asks for the resolution. Empty data keys are filled by you in SmartestQA, never in the chat.

Jira, mailbox and database#

Examples
Is Jira connected for Web Shop? Test the connection, then send defect 45 to Jira.

Wait for the sign-up e-mail in the test mailbox and give me the confirmation link.

Run this read-only query on the TEST database and show the first 10 rows: SELECT …

These use the integrations set up in SmartestQA (Jira, the test mailbox, database connections). A bulk sync and a writing database statement ask for your click.

Report and share#

Examples
Summarise the state of Web Shop for my manager in five bullet points.

Mail the report of the latest Smoke run to qa-lead@example.com.

Mailing a report opens a confirmation dialog with the address.

Asking well#

Do

  • Name the project, and the module or scenario when you know it.
  • Ask for one module at a time.
  • Ask “show me the diff first” when you are unsure.
  • Say what is out of scope: “not the admin pages”.

Don’t

  • Ask it to “test everything” in one go.
  • Paste logins or tokens; let it create empty data keys.
  • Ask it to retry a BLOCKED scenario unchanged.

Didn’t find your answer?Write to support with your project name and what you see on screen.

info@smartestqa.ai