Agents In Use
What’s new todayLast update
108 use cases added today

How to use Cursor to repeat REST bug reproductions

Cursor reproduced authenticated REST failures using Swagger readiness, JWT tooling and curl; Stevens refined its stored workflow until the fourth bug ran alone.

Source published linkedin.comAgent: Cursor
FROM THE SOURCE

Original prompt

Shared by the source author. See source notes for context.

read work-item 1234. The payload for this is 'x' and the endpoint is http://localhost:8080/blah, go fix it

Evidence at a glance

Sourced account
Source author
Gethin Stevens
Original account
Read the source ↗
Published
Source link checked
Reported result
By the fourth service bug, Stevens reports that Cursor reproduced and fixed the issue without intervention; individual bug diffs and root causes are not published.
Human involvement
Stevens supplied the reproduction contract and authentication mechanism, refined the workflow after failures and retained the expected-result decision.

A linked account supports attribution. It does not mean we independently reproduced its result.

Report an issue with this caseOpen correction email

Opens your email app with the case link included. Nothing is sent automatically.

How the agent did it

  1. Stevens supplied a failing payload, local endpoint, expected response and token-acquisition method.
  2. Cursor restarted the service, checked Swagger readiness, authenticated and reproduced the request with curl.
  3. It inspected logs and revised the code until the response matched, while the owner improved the stored workflow after failures.

Gethin Stevens wanted Cursor Debug Mode to reproduce a service failure itself instead of repeatedly asking him to run the code. He already had the REST endpoint and failing input, but the agent also needed the local address and a way to obtain the service’s custom JWT.

He put the operating procedure in a Markdown file. Cursor checked for an existing process, restarted and compiled the service, then waited for Swagger to return HTTP 200. An MCP server supplied an authentication token. Cursor sent the supplied payload with curl, collected logs and compared the response with the expected result. A failure sent it back through hypotheses, instrumentation and another code change.

The procedure itself needed refinement. When a bug-fixing run hit a problem, Stevens asked Cursor to update the workflow so the next run avoided it. By the fourth service bug, he reports that the loop completed without intervention while he worked in a separate Cursor instance.

The article provides a repeatable execution contract, not public diffs or the exact causes of all four bugs. Stevens supplied authentication, inputs and expected behavior, and corrected the workflow. Its demonstrated result is an owner-reported autonomous reproduction-and-fix loop within that prepared service environment.

Prompt
Original prompt

Source: LinkedIn article · Gethin Stevens

Frequently asked questions

How did Cursor authenticate requests?

An MCP server supplied the custom JWT token required by the service.

What prevented requests before startup completed?

The workflow waited for a successful Swagger HTTP 200 response.

Are the individual bug fixes publicly shown?

No. The article documents the operating loop and reported completion, without public diffs.

More details

About this account

Stevens supplied a failing payload, local endpoint, expected response and token-acquisition method. Cursor restarted the service, checked Swagger readiness, authenticated and reproduced the request with curl.

Original source-provided example prompt, with placeholder work item and endpoint; it is not represented as the literal first bug request. Firsthand workflow and exact date opened. Fourth-bug autonomy is owner-reported. Original prompt excerpt.

Credit: Gethin Stevens — LinkedIn article

Source published

Added to Agents In Use

Source checked

Tools and topics

Cursor Debug Mode · curl · JWT authentication MCP · Swagger

bug reproduction · authenticated endpoint · runtime logs

Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.