Capability Map

What this skill does

capability-map produces the product's considered capability set — one README.md per capability — by decomposing product/vision.md. It owns the "is this one capability or three" judgment, and it promotes the provisional capability stubs change-scope created during bootstrap into considered capabilities. It produces the decomposition as a PROPOSAL for human approval; on approval it applies the map, migrating the change history of any capability it reshapes. It is the last pipeline skill.

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 capability map you MUST, in order:

1. CHECK THE VISION. Read product/vision.md. It must exist and be

status: approved. If there is no vision, or it is unapproved — STOP. You cannot decompose a product that has not been defined; run product-vision first.

1. READ THE CURRENT STATE. Read vision.md in full and every existing

capability README.md (provisional and considered) with its changes/ history.

1. DECOMPOSE the product into a considered capability set (see "The

decomposition").

1. PRODUCE THE PROPOSAL — the considered map plus every merge / split / rename

it implies against the current capabilities (see "The proposal").

1. SEEK HUMAN APPROVAL of the proposal. Nothing is rewritten before approval.
2. APPLY the approved map (see "Applying the map"), migrating reshaped

capabilities.

Hard rules:

• No vision, no run. The vision is the criterion you decompose against.
• You never rewrite a README or a why-link before the proposal is approved.
• A capability with shipped changes may be reshaped, never erased. Its history

is real.

The decomposition

Decompose the product — defined by vision.md — into capabilities. A capability is a durable area of the product, named for what it is (watchlists), never for a release or version.

This is a BATCH judgment over the whole product at once, not a stub-by-stub promotion. Holding the vision and the current capabilities together, decide the considered set. You may conclude:

• a provisional stub is sound → promote it as-is;
• two provisional capabilities are really one → merge;
• one stub should be several → split;
• the vision implies a capability nothing has grown toward yet → a new,

empty considered capability.

Each considered capability cites a vision goal — the goal it serves — as its why-link. A capability you cannot tie to a vision goal is a finding: either the vision is incomplete or the capability is misaligned. Surface it in the proposal; do not invent a goal to cite.

Most runs the provisional set will be broadly sound and the decomposition mostly confirms it with light edits. Reserve merges and splits for where the decomposition genuinely changes shape.

Not every thing in the product is a capability. Some are infrastructure — substrate (DESIGN_PRODUCT_PIPELINE.md §2a): authentication, third-party data feeds, the database. When decomposing, classify each such thing explicitly as infrastructure (not a capability): do NOT invent a vision goal to give it a why-link, and do NOT bury it inside one capability — it sits below the capability layer, consumed directly. If a thing looks like infrastructure but the vision implies users experience it as a product surface, it is a capability and the vision has a gap — flag it, do not misclassify it.

The proposal

Produce the decomposition as a PROPOSAL before changing anything. The proposal states:

• the considered capability set, each with its one-line description and the

vision goal it cites;

• for each existing capability, what happens to it — confirmed, merged

into {x}, split into {x, y}, renamed to {x}, or unchanged;

• for every merge / split / rename of a capability that HAS a changes/

history, the migration it entails — which changes/ folders move and which why-links must be rewritten;

• any capability that cannot be tied to a vision goal, flagged for a human.
• any thing classified as infrastructure rather than a capability — listed

separately, with no why-link, noted as below the capability layer.

Present the proposal for review and explicitly invite rejection or revision. A skill that can mass-break why-links must not act unreviewed. Nothing in the state tree is rewritten until a human approves the proposal.

Applying the map

Only after the proposal is human-approved, apply it:

• Confirmed / unchanged capability — write or update its considered

README.md: what the capability is, the vision goal it cites, its boundary. Clear any provisional marker. This is the considered description — capability-map's exclusive output.

• Reshaped capability (merge / split / rename) WITH a changes/ history —

this is a MIGRATION, applied atomically:

1. move the affected changes/{NNNN}-{slug}/ folders to their new

capability;

1. rewrite every why-link and capability: frontmatter field in every

PRD / spec / issues file that referenced the old capability;

1. write the considered README for the resulting capability(ies).

A reshape that skips the migration leaves the tree with broken why-links — it is not done until every link resolves again.

• A capability with shipped changes is never erased. If the decomposition

removes a capability as a standalone, its changes are folded into whichever capability absorbs them — the history moves, it is never dropped. Erasing shipped history would make the repo lie about what was built.

• Infrastructure-tier items — record each in the capability-map output as

infrastructure: below the capability layer, consumed by named capabilities, no why-link, no vision goal. Do NOT create a capabilities/{slug}/ folder or a considered README for it — it is not a capability.

When writing or updating a capability README.md, bump its revision and append a matching ## Changelog entry in the same act — what changed and why (promotion from a provisional stub, a reshape, a reconciliation finding). Revision bump and changelog entry are atomic.

After applying, every why-link in the tree must resolve. Verify it.

capability-map is not the only skill that writes a capability README.md: spec-breakdown updates the present-tense README when a change ships. Any skill that writes a capability README is bound by the same atomic revision+changelog rule (see spec-breakdown's "Keeping the repo true").

When this skill applies

• Mapping a product's capabilities for the first time, or re-mapping them —

this skill.

• Promoting provisional capability stubs into considered capabilities — this

skill.

• No approved vision.md exists — NOT this skill; run product-vision first.
• A single feature idea — NOT this skill; that is change-scope (stage 3).
• Updating one capability's present-tense README because a change shipped —

NOT this skill; that is spec-breakdown.

• Discussion about product structure — not a trigger; answer normally.

Precedence

vision.md outranks capability-map — the vision is the criterion the decomposition serves; this skill never edits the vision to make a capability fit. If the decomposition exposes that the vision is incomplete, that is a finding for product-vision and a human, surfaced in the proposal — not something to resolve by bending the map.

capability-map reads and writes no decision records; technical decisions are below its concern.

What a capability README must contain

A considered capabilities/{slug}/README.md (this skill's exclusive output):

Frontmatter:

---
capability: {slug}
revision: 1
why-link: {resolvable path + goal in product/vision.md}
status: considered
---

Body:

• What this capability is — the durable area of the product it covers,

in present tense.

• Why it exists — the vision goal it serves.
• Boundary — what falls inside this capability and what belongs to a

neighbouring one.

• Changelog — per DESIGN_PRODUCT_PIPELINE.md §2b (Changelog); a

newly-promoted capability's first entry is revision 1 — <date> — capability established.

The README is the considered description. The per-change present-tense detail of what the capability currently does is maintained by spec-breakdown on each ship; capability-map owns the considered framing, not the change-by- change present tense.