Biz Prompt KitSubmit a use case ↗
← All playbooks
Dots: ongoing work / Editorial adaptation

Turn a website problem report into a reviewed fix

Reproduce a customer-facing issue, prepare a focused change and bring back evidence before release.

OpenAI dots + CodexAdvancedOriginal editorial guide

What you’ll produce

A reproducible issue report, proposed code change and relevant test or visual evidence.

Inputs & connections

  • A customer bug report with the affected flow and environment
  • Authorized repository access, project setup and test requirements

A configured Codex cloud environment or an authorized connected computer with the correct repository. Local work requires that computer online with the app open.

Your starting prompt

Replace the bracketed fields. This is our suggested prompt, not a quote from the source.

Investigate [customer website issue] in [authorized repository and test environment]. Reproduce the failure, record the relevant steps and distinguish observed behavior from guesses. Prepare the smallest suitable patch and run checks that exercise the affected flow. Return the diff, results and any remaining uncertainty. Ask for approval before publishing a branch or pull request, merging, deploying, modifying production data or contacting the customer.
Open OpenAI dots + Codex ↗

How to run it

  1. 01

    Confirm the affected flow and reproduce it in the approved environment.

  2. 02

    Trace the cause and prepare a focused local change.

  3. 03

    Run relevant tests and capture evidence of the corrected behavior.

  4. 04

    Review the change and approve publishing or deployment through the normal release process.

KEEP A PERSON IN THE LOOP

Review before you act

A responsible reviewer checks scope, regressions and test evidence before a change reaches customers.

Limitations & success checks

  • A completed agent run is not proof that the issue was fixed or a deployment succeeded.
  • Missing reproduction details or environment differences can prevent a reliable diagnosis.

Measure: Issue reproduced, affected checks passed and reviewer-confirmed resolution after release.

Sources & attribution

The sources below support the reported example or a tool capability. They do not validate our prompt or prove the outcome in your environment.

Official documentationTasks and memory ↗

Official documentation includes a mobile-checkout investigation with a draft PR and test evidence, and explains cloud versus local task requirements. This website workflow is an original adaptation, not an observed customer result.

Creator / publisher: OpenAI

Editorial review: 2026-10-17. How we label evidence.

FROM A USE CASE TO A WORKING SYSTEM

Have a workflow worth building?

Bring a workflow and the tools your team already uses. Ferociter can help scope the project and build the system.

Talk through your project ↗