aiagents.org

Write code that passes tests first

Coding agentKINDA1 method

KINDA — AI can help, with a person in the loop.

1 method on the page: 1 agent.

Use cases

Methods

Write code that passes tests first

agent

Cursor

Steps

  1. Have the agent write the tests, working from the inputs and outputs you expect, and say plainly that you are doing test-driven development so it does not create stand-in versions of code that does not exist yet.
  2. Have the agent run the tests and confirm they fail. Do not let it write feature code yet.
  3. Once you are satisfied with how thorough and sound the tests are, commit them; that locks in the requirements the agent will build to.
  4. Tell the agent to make every test pass without changing the tests, and keep going until all of them pass.
  5. Review the result, check it behaves the way you expect, and commit the code.
  6. The agent can run tests and read failures, so each run gives it concrete feedback; the page says this suits backend code best, where you cannot check by looking at a screen.
Run this method with your agent
Download .mdCopied 0
---
name: "write-code-that-passes-tests-first-method"
description: "Use when you need to Write code that passes tests first. Write code that passes tests first"
---

# Write code that passes tests first

Job: [Write code that passes tests first](https://aiagents.org/canai/write-code-that-passes-tests-first)

## Need

- ai agent: [Cursor](https://aiagents.org/agents/cursor)

## Steps

1. Have the agent write the tests, working from the inputs and outputs you expect, and say plainly that you are doing test-driven development so it does not create stand-in versions of code that does not exist yet.
2. Have the agent run the tests and confirm they fail. Do not let it write feature code yet.
3. Once you are satisfied with how thorough and sound the tests are, commit them; that locks in the requirements the agent will build to.
4. Tell the agent to make every test pass without changing the tests, and keep going until all of them pass.
5. Review the result, check it behaves the way you expect, and commit the code.
6. The agent can run tests and read failures, so each run gives it concrete feedback; the page says this suits backend code best, where you cannot check by looking at a screen.

## Credit

Method author Cursor
Source: https://cursor.com/learn/creating-features

Method author CursorSource