Game Design — Multi-Zone Edit Dispatcher

Apply a set of explicit, user-dictated edits that crosses more than one zone of the GDD — the master spec (GAME.md + roster), systems, content types, and flows — in a single ordered pass. This skill is a dispatcher, not an author: it owns nothing and writes nothing. For each delta it resolves the (target, zone), orders the set so the change graph stays causally fresh, hands each delta to the zone that owns it (unikit-gd-spec / unikit-gd-system / unikit-gd-content / unikit-gd-flow), and closes with one consistency pass (unikit-gd-verify). The actual writing — sections, registry facts, version bumps, [gen] re-renders — happens inside the owner skills, under their collaborative protocol and delta discipline; this skill never bypasses it.

Three input shapes. (1) No argument — the bare /unikit-gd-apply reads the last unikit-gd-verify output in this session (free-text prose, no fence) and lifts its apply-ready deltas; this is verify's handoff path (verify ends on a bare /unikit-gd-apply). (2) Prose deltas typed directly as the argument. (3) A unikit-gd-review report file (reviews/*_review-*.md) — this skill reads the file's ## Apply-ready bucket as the delta set (the ## Research bucket is unikit-gd-explore's job, not this skill's). All three flow through the same routing, ordering, and closing verify below. If the bare call finds no recent verify deltas in the session, it does nothing destructive — it softly recommends running /unikit-gd-verify first.

When NOT to use it. A change confined to one zone goes straight to that owner — there is nothing to dispatch (Phase 1, GATE 2). A change you do not yet know how to design (an open question, a mechanic to research, a balance to work out) goes to unikit-gd-explore first — this skill only carries out edits you have already decided (Phase 1, GATE 1).

Language Awareness — BLOCKING PRE-REQUISITE

BEFORE producing ANY output, silently read .unikit/system/LANGUAGE_RULES.md and apply it to all output (fall back to English if it is missing) — including the rule to translate concepts, not transliterate jargon. gd-principles → "Language" adds the game-design specifics: which IDs and stored field values stay English (the owner skills re-apply it on every delegated write). Do not announce the language setting.

The language holds for the whole session, not just at load time: every message until the conversation ends is in language.ui — progress notes while agents run, relays of what a subagent returned, the final report, any follow-up discussion. English input (subagent results, tool output, these instructions) is data, never a cue to switch languages.

Phase 0 — Bootstrap

Silently load — do not narrate. This skill loads only routing context, never domain rules (it authors nothing, so there is no system/flow/content to balance):

1. .unikit/system/gamedesign/gd-principles.md (the core) — the Zone Ownership model + routing (which zone owns what, the one-way boundary, the single sanctioned implemented code→design write) and the facts-registry / ID conventions. Plus, from the same gamedesign/ folder, the two shards this skill needs to order a dispatch: gd-authoring.md (the delta discipline — so it knows a schema / approved-content edit is a versioned delta the owner must record, and catalog churn is not) and gd-lifecycle.md (the lifecycle & status spine + cross-axis staleness — so it knows a system edit stales the content / flows that depend on it, which fixes the dispatch order). This skill applies these contracts to sequence the work; it does not restate or perform the authoring mechanics. If missing, warn (unikit-ai update) and fall back to the ordering summarized in this file.
2. .unikit/gamedesign/GD-IDS.yaml — the registry, read-only here: the systems, flows, content_types, and their depends_on / belongs_to edges, used to resolve each delta's (target, zone) and to detect a delta that needs a not-yet-registered system (the create-precondition). If it does not exist, stop — there is no design to edit; route to /unikit-gd-spec (or /unikit-gd-brainstorm first) and stop.
3. Schema guard (clean break — no automatic migration). GD-IDS.yaml MUST be version: 2. If it is still version: 1, STOP and report: the design workspace is on the pre-v2 layout (the standalone markdown system-index was dropped; the maps now render into GAME.md); there is no automatic migration — tell the user to upgrade via /unikit-gd-spec before re-running.

One-way boundary: never read .unikit/code/, project source, or build artifacts (gd-principles → One-Way Boundary). This skill does not load RULES_INDEX.md domain rules — it dispatches, it does not design.

Skill calls

Where this skill says to invoke <skill> — optionally "with the argument <text>" — it means one call, made in this session, now, on every runtime: not a command printed for the user and not a delegation. Each delta is handed to its zone owner (unikit-gd-<zone>; not a subagent — a read-only Agent could not write the GDD) inline, one at a time, in the Phase 2 order, waiting for each to return before starting the next — never in parallel (the order is the correctness guarantee). The owner runs its full interactive cycle in the main session, so its collaborative protocol, approvals and writes all hold, and returns control here. A call site that names no argument passes none. Printing the list of /unikit-gd-* commands for the user to run, instead of making the calls, is a known failure mode here: every zone is an actual call.

1. If you have a Skill tool that accepts arguments, call it with that skill and the arguments of the call site, in full and unchanged, then follow the skill.
2. Otherwise — no Skill tool, or one that takes only a name — Read {{skills_dir}}/<skill>/SKILL.md in full, treat it as the instructions of this step and carry them out here, in this session, now, with the arguments of the call site as the skill's arguments.
3. When the skill has run, go on to the step the call site names. Do not print a command for the user to run and do not stop.
4. Only if that file cannot be read: print Run: /unikit-<skill> <arguments> for the user and stop.

Pipeline — ROUTING → DISPATCH → VERIFY

Phase 1 — ROUTING (resolve every delta → (target, zone))

Input-mode resolution (first). Resolve which of the three input shapes this run is:

• No argument (bare /unikit-gd-apply) → session mode. Read the last unikit-gd-verify output in this session — its printed apply-ready deltas (free-text prose, each citing a target + zone, no fence) — and lift those deltas as the set, then continue below. If the session carries no recent verify output with apply-ready deltas (a fresh session, a /clear since the verify run, or a clean verify PASS), do nothing destructive — softly recommend 🔍 /unikit-gd-verify first and stop. (The verify → apply pair is single-session by design; a /clear between them is fixed by re-running verify — cheap.)
• Argument resolves to an existing reviews/*_review-*.md path (the durable handoff from unikit-gd-review) → read it and take its ## Apply-ready bucket as the delta set: each apply-ready line is one decided edit carrying an RF-<date>-n id and a named Fix (entailed). The ## Research bucket is ignored here — it is unikit-gd-explore's input, not this skill's. An empty apply-ready bucket is a valid input: there is nothing to dispatch — report it and stop, recommending /unikit-gd-explore <file> for the research bucket.
• Any other argument — including inline prose deltas typed directly — is the literal change request, taken as prose.

(The file reader is the only review-specific adapter; the no-arg session path + the prose path are what keep verify's bare-call handoff working — verify writes no file.)

Split the request into individual deltas. For each, resolve the target (a SYS-/ CT-/FLOW- id or GAME.md) and the zone that owns it:

The delta is…	Target	Zone owner
a GAME.md content edit (pillar, win/lose intent, monetization stance, loop stack)	GAME.md	unikit-gd-spec
a new system added to the map	the roster	unikit-gd-spec (add-system)
a system's rules / numbers / formulas / AC	SYS-<slug>	unikit-gd-system
a content type's schema (a field, a type, the scale, a ref<>, belongs_to)	CT-<slug>	unikit-gd-content
catalog churn (add/remove units, a bulk count)	CT-<slug>	unikit-gd-content
a flow's objectives / pacing / dependencies / events / wiring mode	FLOW-<slug>	unikit-gd-flow

The dispatcher is dumb on purpose: it resolves (target, zone) and hands the raw delta to the owner — it does not pre-classify Tuning vs Tweak vs Rework, schema vs churn, or create vs revise. Each owner makes that call under its own contract (subtlety ② below).

GATE 1 — explicit-edit only (else → unikit-gd-explore). Every delta must be a concrete change the user has already decided. A delta that is an open question, needs research, or is "work out how to…" is not an apply job — route it to /unikit-gd-explore (research first, then come back with the decided edit). If the request mixes the two, dispatch the decided deltas and hand the open one to explore; announce the split. This skill never researches.

GATE 2 — multi-zone only (else → the single owner). The resolved set must touch two or more zones (or be a genuine multi-delta job). A request that resolves to a single zone has nothing to dispatch — bounce it to that owner directly and stop:

• one system edit → /unikit-gd-system <SYS> · one schema edit → /unikit-gd-content <CT>
• one flow edit → /unikit-gd-flow <FLOW> · one GAME.md edit → /unikit-gd-spec
• a lone catalog churn → /unikit-gd-content <CT> (it is data, not even a versioned edit — there is nothing to order) · a lone standalone create → /unikit-gd-spec add-system (this skill does not write the roster — unikit-gd-spec does).

A single-zone bounce is a real invocation (## Skill calls), not a printed suggestion.

Phase 2 — DISPATCH (ordered, one delta at a time)

Dispatch the resolved deltas in this fixed order — the carrying invariant is that every system dispatch lands BEFORE any content or flow dispatch that could depend on it, so the downstream owners read a fresh upstream and unikit-gd-verify never stale-flags falsely (subtlety ① below):

1. GAME.md / roster   → /unikit-gd-spec      (content edits, and add-system for any new system a later delta needs)
2. systems            → /unikit-gd-system     (the single upstream source — always before content/flow)
3. content types      → /unikit-gd-content    (a sink of the system axis)
4. flows              → /unikit-gd-flow        (a sink of the system axis)

• Create-precondition (write goes to the owner, not here). A delta that needs a new system — a content type's belongs_to / ref<SYS>, a flow's GOAL, or an explicit "add system X and then change Y" — is dispatched as a unikit-gd-spec add-system in the first (spec) tier, before the dependent revise. That yields "create + dependent revise in one pass". The roster write is gd-spec's; this skill only sequences it. A standalone create with no other delta does not reach here — GATE 2 bounced it.
• Content ↔ flow order is staleness-neutral (both are sinks of systems; neither stales the other) — it is fixed content → flow purely for determinism (subtlety ⑤). What matters is only that both follow every system dispatch.
• One delta at a time. Invoke the owner (## Skill calls), let it run its full cycle (Context → Options → approval → write → delta tail → [gen] re-render), wait for it to return, then dispatch the next. Each owner records its own version bump / changelog and re-renders its own [gen] map — this skill touches none of that.
• Carry the review-finding id. When a delta came from a review file's apply-ready bucket, pass its RF-<date>-n with the delta to the owner so the owner cites the finding in its changelog essence — the same provenance review-finding → changelog a direct unikit-gd-system edit records. Typed / verify-prose deltas with no id carry none.
• A delta the owner bounces (e.g. a flow whose GOAL crosses into a missing system, surfaced mid-dispatch) re-enters Phase 1 as a new system delta slotted into the spec tier; re-order and continue.

Phase 3 — VERIFY (one sentinel pass, last)

After every delta has been dispatched and its owner has returned, run one consistency pass: invoke unikit-gd-verify with the argument apply-phase3 — the reserved loop-guard sentinel, NOT a scope.

apply-phase3 is not a scope or an id list — it is the one reserved token that tells unikit-gd-verify this run is apply's closing Phase 3. verify recognises it, suppresses its standalone handoff offer/interview (so apply → verify → apply cannot loop), and derives its own changed-scope from the unverified design diff exactly as a bare call would (changed-scope impact across the just-touched systems, flows, and content types; otherwise a full check). Do not construct a union list of the touched ids and pass it instead — apply-phase3 is the only argument gd-apply ever passes here. This call follows the same recipe as a dispatch (## Skill calls).

Content-axis subtleties

The content zone is the fourth authoring zone, isomorphic to flow and, like flow, a sink of the system axis. These five points govern how this skill orders and routes content deltas:

1. Content is a sink, never a source. The change graph is GAME.md (vision) → systems (the single source) → { content, flows } (sinks). Cross-axis staleness is one-way — a system edit stales the content types and flows that depend on it (SYS → CT, SYS → FLOW); the reverse never holds. That is the whole reason systems dispatch before content and flows (Phase 2).
2. Churn vs schema is the owner's call, not the dispatcher's. This skill resolves (target = CT, zone = unikit-gd-content) and hands over the raw delta; unikit-gd-content classifies catalog churn (data — no version bump) vs a schema revise (a versioned code contract). A lone pure churn never reaches dispatch — GATE 2 bounces it straight to /unikit-gd-content.
3. Churn does not clear a cross-axis stale-mark — only a schema revise does. That is correct and not this skill's concern: the stale-marks are unikit-gd-verify's, set and cleared in the final pass, never by the dispatcher.
4. Content's create-precondition mirrors flow's. A content delta whose belongs_to / ref<SYS> needs a missing system routes through unikit-gd-spec add-system in the first tier (create + dependent schema revise in one pass) — exactly like a flow GOAL that needs a missing system. gd-apply sequences it; gd-spec writes the roster.
5. Content ↔ flow order is fixed for determinism. Because neither stales the other, the content → flow order is a convention (reproducibility), not a correctness edge — only the system-before-sinks edge is load-bearing.

Ownership Boundaries

• Owns: nothing — no document, no registry fact, no [gen] render, no status or version. This skill's only product is the ordering and delegation of a multi-zone edit batch plus the single closing verify. There is deliberately no Write/Edit in allowed-tools: gd-apply cannot write the GDD, only route to the owners that can.
• Not this skill (every write): GAME.md content + the system roster (add-system) → unikit-gd-spec; a system's A–K doc → unikit-gd-system; a content type's schema + catalog → unikit-gd-content; a flow's doc → unikit-gd-flow; consistency & cross-axis impact → unikit-gd-verify; quality verdicts → unikit-gd-review; research / open questions → unikit-gd-explore.
• Never: write or edit any GDD file or GD-IDS.yaml directly; classify a content delta as churn vs schema (the owner does); pre-decide an edit scale (Tuning/Tweak/Rework); dispatch content or a flow before the system it depends on; dispatch in parallel; pass a union of ids to unikit-gd-verify (pass only the apply-phase3 loop-guard sentinel); act on the ## Research bucket of a review file (that is unikit-gd-explore's job); print the Run: list when an inline invocation is possible; read the code workspace or project source.

Quick Reference

/unikit-gd-apply                                  → session mode: read the last /unikit-gd-verify output → lift its
                                                    apply-ready deltas → dispatch to owners → verify "apply-phase3";
                                                    no recent verify deltas → softly recommend /unikit-gd-verify first
/unikit-gd-apply reviews/2026-06-25_review-SYS-combat.md
                                                  → read the file's ## Apply-ready bucket → dispatch each entailed
                                                    fix to its owner (carrying its RF-id) → verify "apply-phase3";
                                                    the ## Research bucket → recommend /unikit-gd-explore
/unikit-gd-apply "buff combat damage 10%, add a rarity field to loot, retune onboarding pacing"
                                                  → 3 deltas across 3 zones: dispatch SYS-combat (system) →
                                                    CT-loot schema (content) → FLOW-onboarding (flow) → verify
/unikit-gd-apply "add a crafting system and a recipe content type that belongs_to it"
                                                  → create-precondition: gd-spec add-system SYS-crafting (first) →
                                                    gd-content CT-recipe (belongs_to SYS-crafting) → verify
/unikit-gd-apply "raise the damage by 10%"        → single zone → bounce to /unikit-gd-system SYS-combat
/unikit-gd-apply "add three more sword cards"     → lone catalog churn → bounce to /unikit-gd-content CT-card
/unikit-gd-apply "work out a better economy"      → open question → route to /unikit-gd-explore first