# Requirements Process for ChatGPT

Version 0.2.0-alpha.4. Tested alpha release, October 25, 2026.

## What this is

Requirements Process is a skill distributed in a ChatGPT plugin. It guides you through Clarify the assignment, Organize the areas, and Specify the requirements. The approved graphic, AI collaboration guidance, maintained dashboard, full-state PDF exporter and private-cloud setup resources are included.

## Install

1. Download `requirements-process-plugin-0.2.0-alpha.4.zip`. Earlier 0.1.0 archives are superseded candidates.
2. In ChatGPT Plugins, choose **Add → Upload plugin archive** and select the ZIP.
3. Add Requirements Process, then install it. For an existing installation, use its **More actions → Upload new version** flow instead of creating a duplicate.
4. Select **Try in chat**, use **Work**, and say: “Use Requirements Process to help me with [assignment].”
5. The skill introduces the method and first asks for your assignment. Once identified, it resumes relevant records or creates a new assignment without overwriting earlier work.

## Private dashboard setup

Use ChatGPT Work with Sites, D1 and MCP provisioning available. Ask the skill to set up your private requirements workspace. It uses the bundled production source to provision your own private Site and companion chat-tool plugin. Follow the supported connection flow if prompted. Installing the workflow ZIP alone does not establish a hosted workspace.

Keep the generated workspace identity and use it when resuming in another conversation. Each state operation names the assignment and expected saved version. Your browser view pins the assignment; changing another tab's selection does not silently redirect it. This package contains no author's account connection, private URL, database or fictional test records.

Keep the dashboard open. It reflects saved progress automatically, so you do not have to ask for a refresh. Chat remains the place to develop and refine your thinking. Display controls do not initiate AI work. Initial stage-three review also offers Keep / Change / Remove; detailed versions offer Confirm specification / Needs more work.

## What the statuses mean

- Keep completes the initial review choice; it is retained in history after that review is settled.
- Open means the detailed specification still needs work or acceptance.
- Confirmed means you accepted what the requirement must include, do or satisfy.
- Ready means its applicable execution prerequisites are met with evidence and you accepted that assessment.
- Done means all active requirements are Ready, the whole-set review is current, and responsibility and handoff choices are accepted.

A completed readiness review may conclude that work remains; it does not automatically produce Ready. Meaningful changes require the relevant acceptance again. None of these labels authorizes spending, contracting, publication or execution. Complex projects still need detailed project planning.

## PDF and handoff

Print / PDF is available throughout the process. It includes the complete dated saved state, histories, sources, next steps and readiness findings regardless of collapsed sections. Unfinished work is labeled. The optional person view groups Person → Area → Requirement → Next activity, with unassigned work separate. Decide during readiness whether named responsibility is required; the final export does not invent owners.

## Verified support and remaining limits

The authorized private ChatGPT account has been used to verify plugin installation, hosted dashboard persistence after reload, actual companion-tool read/write, a second clean assignment, and downloaded PDF content. Independently rerun checks: 301 cloud assertions and 26 local tests, all passing. The live test records are fictional and were not marked Confirmed or Ready for release verification.

A fresh other-account installation and physical phone/iPad remain untested. Sites availability and workspace policies can vary; this release is not a promise of support on every ChatGPT plan. The workflow must report an unavailable capability rather than pretend a connection exists. Claude has a separate tested alpha release and installation guide. The optional desktop path requires Python and ReportLab; the hosted Worker does not require Python.