Change Scope

What this skill does

change-scope produces a PRD — the WHY of one change to a capability: the problem, the user's current pain, the product value, user stories, and scope. It runs BEFORE any technical spec. It owns the product reasoning and makes no technical decisions. The PRD it produces is the required input to change-spec, and change-spec will refuse a PRD that a human has not approved.

Where the state tree lives

Generic skill — hard-codes no path. Resolve the state-tree root from the project's CLAUDE.md (project-declared paths, DESIGN_PRODUCT_PIPELINE.md §2); default docs/, and say which you used. {capabilities-root} means {declared-root}/capabilities (default docs/capabilities), and product/vision.md means {declared-root}/product/vision.md.

STOP — how this skill works

Before producing a PRD you MUST, in order:

1. CLASSIFY the idea. State the classification and the reasoning out loud:

• New capability — the product has no existing area for this.
• Change to an existing capability — name the capability. Make this call

ONLY after reading that capability's existing PRDs and decision records (see Precedence); prior decisions are often what make a change architectural.

• In-shape work — a small modification needing no PRD and no spec. Say so

and STOP: direct the work to the relevant rules skill's code gate. Do not manufacture a PRD for a one-line change.

1. RESOLVE THE CAPABILITY ANCHOR.

• The capability already exists → the PRD will cite its README.md.
• The capability does NOT exist → create a provisional capability stub

(see "Creating a provisional capability stub") so the PRD has a real parent to cite. The stub is an anchor, not a capability description.

1. ALLOCATE THE CHANGE NUMBER (see "Numbering a change").
2. WRITE THE PRD (see "What a PRD must contain").
3. SEEK HUMAN APPROVAL (see "Approval"). Until a human approves, the PRD is not

approved and change-spec may not consume it.

Hard rules:

• You make no technical decisions. No class shapes, casts, schema, or

framework specifics appear in a PRD. If you are writing one, stop — that is change-spec's job.

• A good PRD passes this test: a non-engineer can read it and disagree with

it. If half of it is only meaningful to someone who has read the code conventions, the why and the how have been welded — start over.

• You do not approve your own PRD. Approval is an explicit human act (see

"Approval").

Creating a provisional capability stub

When the target capability does not yet exist, create — and nothing more:

• the folder {capabilities-root}/{capability-slug}/
• a README.md containing ONLY: the capability name, a one-line

description, and this status block:

status: provisional — established by change-scope, pending capability-map why-link: BOOTSTRAP — no product vision yet

If product/vision.md already exists, replace the why-link line with a real citation to the relevant vision goal (path + section) instead of the BOOTSTRAP token.

This stub is a coat hook: it exists so the PRD has a resolvable parent. You MUST NOT write a considered capability description, scope, boundaries, or decomposition into it — that is capability-map's exclusive output, produced later, which will review and promote this stub. Creating the stub does not make change-scope a two-transition skill: the stub is the PRD's parent coming into existence as a side effect, explicitly marked provisional.

Numbering a change

Scan the capability's changes/ directory and take the next free integer, zero-padded to four digits (0001, 0002, …). This number is PROVISIONAL until merge — tell the user plainly: "scoped as NNNN — provisional; if NNNN already exists when this merges, renumber."

The change's identity is its SLUG, not its number. The change folder is changes/{NNNN}-{change-slug}/; the slug is what is unique and meaningful. Two developers minting the same number in parallel is a loud, merge-time naming conflict resolved by a rename — it is expected and acceptable.

Because the number is provisional, every why-link that references THIS change (the spec citing this PRD, issues citing this change) MUST cite the change SLUG, never the bare number. A renumber must never break a why-link.

What a PRD must contain

Write the PRD to {capabilities-root}/{capability-slug}/changes/{NNNN}-{slug}/prd.md.

Frontmatter:

---
change: {NNNN}-{slug}
capability: {capability-slug}
why-link: {resolvable path + section of the capability README.md}
status: draft
approved_by:
approved_on:
---

Body — every section is required; sections collapse for a small change but are never absent:

• Problem — what the user cannot do today, and the pain it causes. In

product terms, no implementation language.

• Why now / value — what shipping this is worth, to whom.
• User stories — real user-facing stories. "As a developer, I want a typed

data contract" is NOT a user story — it is a technical decision; it does not belong here.

• Scope — what this change includes.
• Out of scope — what it explicitly does not, and why.
• Open product questions — unresolved product (not technical) questions.

The PRD describes intent and value only. It names no classes, tables, casts, routes, or framework mechanics.

Approval

A PRD is approved only by an explicit human act. You cannot approve it; a sub-agent cannot; ambiguous assent ("looks good", "sure") does not count — if the response is vague, ask again.

Steps:

1. Present the finished PRD to the user FOR REVIEW. Surface what it commits to

— the problem framing, the scope, the out-of-scope calls — and explicitly invite rejection or revision. Do not present it as a fait accompli; an approval step that can only produce "yes" is theatre.

1. On unambiguous human approval, write into the PRD frontmatter:

status: approved, approved_by: {who}, approved_on: {date}.

1. Any later edit to an approved PRD INVALIDATES the approval: reset

status to draft and clear approved_by/approved_on. The edited PRD needs fresh human approval. "Approved" must always mean "approved as it now reads."

change-spec's STOP gate checks exactly this: it consumes a PRD only if status: approved and the approval fields are populated and current.

Precedence

A project's documented decisions outrank this skill. Read them if the project keeps them — change-scope never requires them and never hard-codes their location; the project's CLAUDE.md declares which exist (commonly docs/adr/, docs/postmortem/, docs/learnings/).

When classifying a change to an existing capability (STOP step 1), read the capability's existing PRDs AND any relevant decision records — read the document body, never the title or a commit message. Prior decisions are often what make an otherwise-small change architectural, and that classification is the difference between scoping it properly and routing it to the code gate.

change-scope proposes no decision records and overrides none. A change that contradicts an existing decision is surfaced for a human ruling in the PRD's Open product questions; the technical resolution belongs to change-spec.