# Requirements Process for Claude

Tested alpha: 0.3.0-alpha. October 25, 2026.

## Install

1. Download `requirements-process-claude-0.3.0-alpha.skill` (the `.zip` contains the identical payload).
2. Open **Claude → Customize → Skills → Yours → Add skill → Upload skill**.
3. Choose the file and select **Upload**. Claude recognizes the skill as `requirements-dashboard`.
4. Verify that it is enabled. Select **Try in chat**, or ask: “Use Requirements Process to help me with [assignment].”

This import path was verified in the signed-in personal Claude Pro account. It is Claude's custom-skill installation flow, not a ChatGPT plugin import. [Claude skill instructions](https://support.claude.com/en/articles/12512180-use-skills-in-claude).

## Begin or resume

The skill explains the three stages and presents the approved overview: Clarify the assignment, Organize the areas, Specify the requirements. Describe what you want to accomplish, where you are now and what you already have. Challenge or redirect the AI whenever a question or proposal is unhelpful.

After mapping your description, the skill creates or resumes your own private Requirements workspace. Installing the skill alone does not create that workspace. This maintained version needs Claude's Artifact and ArtifactData tools. If they are unavailable, the skill must explain the gap and offer an explicitly labeled snapshot rather than pretend it has a live connection.

Claude labels artifact creation “Publish,” but the creation dialog states that a new artifact stays private until shared. Verify **Only you** or **private** after creation. Public sharing is not required for this workflow. Current Claude artifacts support private persistent storage without public publication. [Claude artifact documentation](https://support.claude.com/en/articles/17153992-what-are-artifacts-and-how-do-i-use-them).

Keep the dashboard open. Chat and the page use the same saved record; saved changes update the page automatically. Each assignment has its own record. Resume through the existing workspace and relevant assignment, preserving decisions and history.

## Review and decide

The first two stages use chat for decisions. Stage three first offers Keep / Change / Remove across the proposed requirements. Once that review is settled, the detailed specifications remain Open until you accept them as Confirmed. Readiness work records prerequisites, evidence, progress and responsibility; only an accepted, satisfied assessment makes a requirement Ready. Done additionally requires a current whole-set review and final handoff acceptance.

**Use one writer at a time.** While reviewing choices on the page, let a save finish before asking Claude to change that record. Chat writes reject stale store versions. Page writes use a lease and version check, with a small race window against a simultaneous chat save. The skill reads before and after chat changes; this is not a claim of fully atomic page/chat synchronization.

## PDF and handoff

Both PDF buttons download the full saved state, independent of collapsed sections. Accept Claude's **Save file** prompt. The person view groups next activities by Person → Area → Requirement; unassigned work remains explicit. PDFs can be exported before completion and are labeled work in progress.

The PDF includes the exact UTF-8 saved record as `complete-saved-state.json` and an escaped JSON appendix. Built-in PDF fonts have limits: some characters are visibly transliterated or shown as Unicode codes. The exact original text remains in the attachment and appendix. The export does not change decisions.

Ready and Done do not authorize spending, contacting people, publishing, booking or execution. Complex work still needs detailed project planning.

## Verified support and limits

Verified: private native page/chat round trip, two separate assignments, consecutive page review saves and reload, actual current-state/person PDF downloads, exact attached state, installation, enabled status, visible onboarding image and read-only resume from a fresh conversation. The developer browser suite uses a local stand-in for Claude's documented APIs; its result is distinguished from the live tests in the audit.

A separate account and physical phone/iPad remain untested. A 390-pixel automated browser test is not a physical-device check. Availability on other Claude plans, Cowork or other surfaces must be verified before claiming support there. The ChatGPT release remains separate and unchanged.