@ankhorage/devtools 1.9.5 → 1.10.0

package/dist/tools/skills/assets/ankhorage-project-structure/references/skill-distribution.md

@@ -54,6 +54,11 @@ Do not copy every available skill into every repository.
 - External, personal, experimental, and task-specific skills are not part of Devtools sync.
 - A repository may own additional skill directories beside the Devtools-managed set.
 
+The `zora-designer` profile skill is selected for the ZORA, Templates, and Studio owners and for
+generated-app authoring repositories that consume ZORA with the Ankhorage Runtime, Templates, or
+Studio application model. Plain libraries and unrelated tooling packages retain only baseline
+skills.
+
 Prefer the same repository-trait detection used by other Devtools profiles. Add explicit metadata
 only for an actual ambiguity; do not require each repository to repeat a default skill list.
 

package/dist/tools/skills/assets/zora-designer/SKILL.md

@@ -0,0 +1,89 @@
+---
+name: zora-designer
+description: >
+  Design one application screen or an ordered screen series, audit a URL or supplied image
+  evidence, and author an Ankhorage Templates manifest or starter using installed ZORA metadata
+  and theme APIs. Use for category-driven design decisions, design audits and scoring, visual
+  reconstruction with ZORA elements, or template creation. Do not use for illustration-only work.
+---
+
+# ZORA Designer
+
+Design, audit, and author through the target repository's released owner APIs. Keep
+`zora-designer.md` as review evidence; the application manifest and owner contracts remain runtime
+authority.
+
+## Route the request
+
+- `interactive`: resolve a design configuration without silently creating code or images.
+- `screen`: design one screen and, when requested, produce a concept image or implementation.
+- `screens`: design a coherent ordered series with shared navigation, state, and tokens.
+- `audit`: evaluate a URL, one supplied image, or an ordered image series with explicit evidence.
+- `template`: compose a canonical `AppManifest`; in Templates, scaffold the normal starter source
+  and registry entry.
+
+Natural language is enough. Ask only when an unresolved choice would materially change the output.
+Read [workflow.md](references/workflow.md) for the ordered process and capability gates.
+
+## Start with the owners
+
+From the target repository, run:
+
+```text
+bun .agents/skills/zora-designer/scripts/owner-api.mjs inspect
+```
+
+The helper loads public exports from installed `@ankhorage/templates`, `@ankhorage/zora/theme`, and
+`@ankhorage/zora/metadata`. If it reports a missing or outdated owner, update the released package
+through the repository's normal dependency workflow. Never copy a catalog, token inventory, color
+algorithm, theme compiler, component schema, or manifest implementation into this skill.
+
+Use owner terminology verbatim: `GeneratedColorRole`, `ThemeTokens.colors`, `ThemeSemantics`, and
+`SemanticColorToken`. Do not introduce aliases or naming cleanup here.
+
+## Preserve decisions and evidence
+
+Resolve values in this order: current request, current session, verified project state, confirmed
+existing brief target, category preset, then safe global default. Record an origin for each value.
+Keep target design separate from observed runtime state and report drift rather than overwriting one
+with the other.
+
+For scoring, evidence limits, release gates, and finding rules, read
+[audit.md](references/audit.md). For deterministic `zora-designer.md` shape and serialization, read
+[artifact.md](references/artifact.md).
+
+## Compose only supported ZORA elements
+
+Inspect `ZORA_COMPONENT_META` and `ZORA_THEME_RECIPE_META`. Map a source region only when the
+element's semantic responsibility, structure, states/interactions, accessibility contract, props,
+events, and data requirements fit. Visual resemblance alone is insufficient.
+
+If no exact element fits:
+
+1. use the metadata-backed `MissingElement` draft node at that location;
+2. record the evidence, warning, requested capability, and application blocker;
+3. create or link a ZORA owner issue;
+4. stop release application until a released real element replaces it.
+
+Never disguise the gap with a generic container, invented prop, or custom application workaround.
+
+## Capability truthfulness
+
+- Concept image output requires an available image-generation capability.
+- URL intake and authoritative runtime screenshots require a browser/runtime capture capability.
+- Image audits require the supplied original image or ordered series.
+
+When a required capability or source is unavailable, report that deliverable or evidence scope as
+blocked. Do not simulate an image, browser observation, or runtime proof. Clearly distinguish
+generated concept images from authoritative runtime captures.
+
+## Validate before handoff
+
+- Compile the category design and both theme modes through installed owner APIs.
+- Validate manifest composition and preserve every owner diagnostic.
+- Confirm every manifest node and recipe against current ZORA metadata.
+- Run deterministic audit arithmetic when an audit is in scope.
+- Keep screenshot-only behavior `not-assessable`.
+- Require a ready application gate before release; a score never overrides a blocker.
+- Update `zora-designer.md` and report target/runtime drift, unsupported capabilities, and owner
+  issues.

package/dist/tools/skills/assets/zora-designer/agents/openai.yaml

@@ -0,0 +1,6 @@
+interface:
+  display_name: 'ZORA Designer'
+  short_description: 'Design, audit, and author Ankhorage screens and templates through ZORA.'
+  default_prompt: 'Use $zora-designer to design, audit, or author this Ankhorage interface.'
+policy:
+  allow_implicit_invocation: true

package/dist/tools/skills/assets/zora-designer/assets/audit-rubric.json

@@ -0,0 +1,312 @@
+{
+  "schema": "zora-designer/audit-rubric/v1",
+  "rules": [
+    {
+      "id": "R01",
+      "name": "Clarity",
+      "weight": 7,
+      "generation": "Use unambiguous labels, recognizable controls, plain-language instructions, and a visible primary task.",
+      "audit": "Check calls to action, icons, headings, affordances, state, and consequences; do not infer accessible names from pixels.",
+      "criteria": [
+        "purpose-and-primary-task",
+        "labels-and-copy",
+        "icons-and-affordances",
+        "consequences-and-state-clarity"
+      ]
+    },
+    {
+      "id": "R02",
+      "name": "Consistency",
+      "weight": 5,
+      "generation": "Reuse owner tokens, components, terminology, navigation placement, and state behavior.",
+      "audit": "Group repeated one-off styling, labels, placement, or behavior under their systemic root cause.",
+      "criteria": [
+        "tokens-and-components",
+        "terminology",
+        "navigation-and-placement",
+        "interaction-and-state-behavior"
+      ]
+    },
+    {
+      "id": "R03",
+      "name": "Simplicity",
+      "weight": 4,
+      "generation": "Keep elements that support comprehension, completion, confidence, or brand intent; disclose advanced detail progressively.",
+      "audit": "Check redundant controls, avoidable steps, repeated explanation, and decorative competition without penalizing required domain complexity.",
+      "criteria": [
+        "necessity-and-redundancy",
+        "task-step-efficiency",
+        "progressive-disclosure",
+        "decorative-competition"
+      ]
+    },
+    {
+      "id": "R04",
+      "name": "Visual hierarchy",
+      "weight": 7,
+      "generation": "Create a deliberate reading and action order through scale, weight, contrast, position, grouping, and whitespace.",
+      "audit": "Compare focal priority, heading order, action emphasis, and visual order against semantic and focus order.",
+      "criteria": [
+        "focal-priority",
+        "heading-and-content-order",
+        "action-priority",
+        "visual-semantic-focus-order"
+      ]
+    },
+    {
+      "id": "R05",
+      "name": "Scannability",
+      "weight": 5,
+      "generation": "Chunk content with meaningful headings, concise labels, stable alignment, and task-appropriate line length and density.",
+      "audit": "Check walls of text, weak anchors, long forms, comparison alignment, and key-information discovery.",
+      "criteria": [
+        "chunking-and-headings",
+        "alignment-and-comparison",
+        "line-length-and-density",
+        "key-information-discovery"
+      ]
+    },
+    {
+      "id": "R06",
+      "name": "Color contrast",
+      "weight": 10,
+      "generation": "Meet applicable contrast requirements in every state, mode, overlay, and composite background using computed owner values.",
+      "audit": "Use measurements for ratios; visual evidence can identify risk but cannot prove exact composited contrast.",
+      "criteria": [
+        "normal-and-large-text",
+        "active-ui-and-graphics",
+        "focus-and-selection",
+        "states-overlays-images-gradients"
+      ]
+    },
+    {
+      "id": "R07",
+      "name": "Intentional color",
+      "weight": 5,
+      "generation": "Assign color through semantic roles, use high chroma sparingly, and provide a non-color cue for every meaning.",
+      "audit": "Check semantic conflicts, color-only state, competing accents, and separation of brand from severity meaning.",
+      "criteria": [
+        "semantic-role-consistency",
+        "non-color-cues",
+        "accent-restraint",
+        "brand-status-separation"
+      ]
+    },
+    {
+      "id": "R08",
+      "name": "Negative space",
+      "weight": 4,
+      "generation": "Use whitespace to reveal grouping, hierarchy, focus, and rhythm at the intended product density.",
+      "audit": "Check control crowding, edge collisions, uneven rhythm, disconnected empty regions, and space masking weak hierarchy.",
+      "criteria": [
+        "internal-rhythm",
+        "external-rhythm",
+        "edge-clearance-and-group-separation",
+        "density-balance"
+      ]
+    },
+    {
+      "id": "R09",
+      "name": "Typography hierarchy",
+      "weight": 7,
+      "generation": "Use a small tokenized type scale, clear role mapping, readable metrics, verified fonts, and robust scaling.",
+      "audit": "Check arbitrary styles, essential text size, line height, unavailable weights or glyphs, clipping, and semantic heading alignment.",
+      "criteria": [
+        "roles-and-scale",
+        "line-height-measure-tracking",
+        "font-glyph-availability",
+        "scaling-clipping-semantics"
+      ]
+    },
+    {
+      "id": "R10",
+      "name": "Immediate feedback",
+      "weight": 5,
+      "generation": "Define relevant pressed, selected, hover, focus, loading, success, error, empty, and disabled behavior.",
+      "audit": "Check silent actions, unstable progress, duplicate submission, and state feedback; static images cannot pass behavior criteria alone.",
+      "criteria": [
+        "direct-interaction-feedback",
+        "async-loading-progress",
+        "outcome-and-empty-feedback",
+        "focus-hover-disabled-states"
+      ]
+    },
+    {
+      "id": "R11",
+      "name": "User control",
+      "weight": 5,
+      "generation": "Provide exit, cancel, undo, pause, retry, and reversible paths proportional to risk while preserving user work.",
+      "audit": "Check traps, unnecessary irreversibility, destructive action safety, and dismissals that lose data or context.",
+      "criteria": [
+        "exit-cancel-back",
+        "undo-retry-reversibility",
+        "destructive-confirmation",
+        "input-and-context-preservation"
+      ]
+    },
+    {
+      "id": "R12",
+      "name": "Error prevention and recovery",
+      "weight": 7,
+      "generation": "Prevent predictable errors, validate at useful times, identify the problem and remedy, and preserve valid input.",
+      "audit": "Check constraints, defaults, timing, error association, human-readable recovery, retry, and data preservation.",
+      "criteria": [
+        "prevention-constraints-defaults",
+        "validation-timing",
+        "error-identification-association",
+        "recovery-and-data-preservation"
+      ]
+    },
+    {
+      "id": "R13",
+      "name": "Familiar patterns",
+      "weight": 4,
+      "generation": "Use platform and domain conventions unless a different pattern is demonstrably more suitable and learnable.",
+      "audit": "Check surprising navigation, placement, or control behavior and familiar-looking elements that behave unexpectedly.",
+      "criteria": [
+        "navigation-conventions",
+        "control-conventions",
+        "domain-flow-conventions",
+        "innovation-learnability"
+      ]
+    },
+    {
+      "id": "R14",
+      "name": "Choice limitation",
+      "weight": 3,
+      "generation": "Prioritize choices, provide defaults, group options, and progressively reveal advanced decisions.",
+      "audit": "Check equal-priority action clusters, ungrouped menus, deferrable decisions, and large sets without discovery support.",
+      "criteria": [
+        "priority-and-defaults",
+        "option-grouping",
+        "progressive-choice-disclosure",
+        "search-filter-for-large-sets"
+      ]
+    },
+    {
+      "id": "R15",
+      "name": "Target sizing",
+      "weight": 7,
+      "generation": "Size and space complete interactive bounds for the actual touch, pointer, keyboard, and assistive input profile.",
+      "audit": "Measure bounds rather than glyphs and consider platform baselines, exceptions, edge placement, reach, frequency, and consequence.",
+      "criteria": [
+        "interactive-bound-size",
+        "spacing-and-overlap",
+        "reach-edge-frequency",
+        "input-specific-equivalence"
+      ]
+    },
+    {
+      "id": "R16",
+      "name": "Proximity and grouping",
+      "weight": 5,
+      "generation": "Place labels, help, errors, actions, and content near the items they describe and expose groups without color alone.",
+      "audit": "Check detached labels or errors, misassociated actions, ambiguous containers, and grouping based only on color.",
+      "criteria": [
+        "label-control-association",
+        "help-error-proximity",
+        "action-container-association",
+        "group-boundary-and-alignment"
+      ]
+    },
+    {
+      "id": "R17",
+      "name": "Accessibility",
+      "weight": 10,
+      "generation": "Support semantics, names, keyboard and focus, alternatives, reflow, scaling, orientation, reduced motion, localization, and assistive updates.",
+      "audit": "Use appropriate runtime, source, and accessibility evidence; score contrast under R06 and target size under R15 without duplicate deductions.",
+      "criteria": [
+        "semantics-names-roles-values",
+        "keyboard-focus-order-and-traps",
+        "reflow-zoom-scaling-orientation",
+        "alternatives-announcements-motion-localization"
+      ]
+    }
+  ],
+  "statusFactors": {
+    "pass": 1,
+    "minor": 0.75,
+    "major": 0.4,
+    "critical": 0,
+    "not-assessable": null,
+    "not-applicable": null
+  },
+  "confidenceFactors": {
+    "high": 1,
+    "medium": 0.67,
+    "low": 0.33,
+    "unknown": 0
+  },
+  "releaseGates": [
+    {
+      "id": "RG01",
+      "name": "core-task-operability",
+      "criteria": [
+        "primary-task-reachable-and-completable",
+        "required-controls-operable",
+        "state-and-outcome-perceivable",
+        "no-blocking-task-failure"
+      ]
+    },
+    {
+      "id": "RG02",
+      "name": "keyboard-focus-and-no-trap",
+      "criteria": [
+        "keyboard-equivalent-operation",
+        "visible-focus",
+        "logical-focus-order",
+        "no-keyboard-trap-and-contained-modal-focus"
+      ]
+    },
+    {
+      "id": "RG03",
+      "name": "semantics-names-roles-values-and-announcements",
+      "criteria": [
+        "semantic-structure",
+        "accessible-names",
+        "roles-states-and-values",
+        "announcements-errors-and-live-updates"
+      ]
+    },
+    {
+      "id": "RG04",
+      "name": "visual-AA-contrast-color-reflow-and-resize",
+      "criteria": [
+        "text-contrast",
+        "non-text-contrast-and-color-cues",
+        "reflow-zoom-orientation-and-text-spacing",
+        "text-resize-without-clipping-or-loss"
+      ]
+    },
+    {
+      "id": "RG05",
+      "name": "input-AA-target-drag-and-gesture-operation",
+      "criteria": [
+        "minimum-target-size-or-exception",
+        "target-spacing-and-obstruction",
+        "drag-and-complex-gesture-alternative",
+        "pointer-touch-and-motion-equivalence"
+      ]
+    },
+    {
+      "id": "RG06",
+      "name": "error-recovery-and-destructive-safety",
+      "criteria": [
+        "error-identification-and-prevention",
+        "recovery-and-input-preservation",
+        "destructive-confirmation-or-reversal",
+        "cancel-back-retry-and-context-preservation"
+      ]
+    },
+    {
+      "id": "RG07",
+      "name": "blocking-owner-or-runtime-diagnostics",
+      "criteria": [
+        "owner-diagnostics-pass",
+        "runtime-matches-approved-target",
+        "no-unresolved-application-blocker",
+        "required-validation-evidence-present"
+      ]
+    }
+  ]
+}

package/dist/tools/skills/assets/zora-designer/references/artifact.md

@@ -0,0 +1,93 @@
+# Portable `zora-designer.md` Artifact
+
+`zora-designer.md` is the deterministic design/audit decision record. It is never runtime theme or
+manifest authority. The serializer and its ordered frontmatter/section skeleton live in
+[audit.mjs](../scripts/audit.mjs); do not maintain another template.
+
+## Document lifecycle
+
+Use these document statuses:
+
+- `draft`: required decisions or evidence are still unresolved;
+- `resolved`: the requested configuration is complete and applicable;
+- `resolved-with-gaps`: design decisions are complete but owner/runtime support is not;
+- `blocked`: the requested deliverable cannot proceed;
+- `audited`: the available evidence has been assessed and limitations are explicit.
+
+Use `documentKind: configuration` for interactive/design/template work and `audit` for an audit.
+Generation self-checks belong under validation; do not create a pretend audit.
+
+## Canonical input to the serializer
+
+Provide a JSON object with these responsibilities. Missing optional data stays empty or `null`; it
+is never invented:
+
+- `language` and `status`;
+- `source`: mode, original inputs, ordered evidence, capability limitations;
+- `config`: requested/resolved category, intent, platform target profiles, theme coverage, primary,
+  harmony, tone pair, typography, and advanced profile, each with origins;
+- `derivation`: owner provenance, diagnostics, assumptions, unsupported concepts, and explicit
+  target/runtime drift;
+- `tokens`: current owner-computed token output, never copied token definitions;
+- `components`: metadata-backed recipe decisions and required states;
+- `screens`: ordered screen specifications and evidence relationships;
+- `validation`: scope, gates, application gate, owner/runtime drift, and blockers;
+- `auditInput`: criterion and release-gate assessments consumed by the canonical calculator;
+- `findings`, `risks`, `openDecisions`, and preserved `userNotes`.
+
+The serializer emits a JSON-compatible YAML frontmatter document with stable key order and a fixed
+human-readable section order. JSON is a valid YAML subset and avoids ambiguous scalar coercion.
+Never add timestamps, random choices, environment paths, or unstable prose.
+
+## Origin and target/observed rules
+
+Every resolved input records one origin from:
+
+```text
+user | session | project | existing-brief | category-default | global-default |
+observed | inferred | derived | unknown
+```
+
+Current explicit input wins. `derived` means mechanically resolved from separately recorded inputs.
+Preserve user notes verbatim. Regenerate derived content rather than retaining obsolete generated
+values as compatibility state.
+
+During audits and migrations, declared target configuration and observed implementation remain
+separate. Record owner/runtime drift with owner, path, owner value, design target, evidence, failed
+gate, and required owner change. A complete target can be `resolved-with-gaps` while application is
+blocked.
+
+## Validation states
+
+Validation scopes are `configuration`, `composition`, `runtime`, and `audit`. A gate is
+`not-applicable`, `not-assessable`, `pass`, `fail`, or `blocked`. Configuration-only work does not
+fail for absent runtime interaction evidence; it marks that scope not applicable.
+
+Aggregate validation in this order: blocker, failure, not run, pass with non-blocking gaps, then
+pass. Resolve the application gate separately. A runtime blocker affects aggregate status only when
+runtime application is in the requested scope.
+
+Every unresolved `MissingElement` records a blocker, owner issue link when one exists, and the
+condition for replacing it with a released real ZORA element.
+
+## Deterministic audit fields
+
+The audit calculator supplies score, coverage, applicable and assessed weight, rounding policy,
+confidence, possible range, release gate and items, complete rule results, findings with allocated
+score impact, risks, passed rules, and not-assessable criteria. Keep unavailable scalar values as
+`null`; do not omit them or substitute zero.
+
+An image or series records each original source separately with dimensions and order. A URL records
+the capture viewport, theme mode, state, and reproduction. Concept images are labeled generated;
+runtime captures are labeled observed/measured and include their capture source.
+
+## Output and persistence
+
+Write to an explicit output path when supplied; otherwise update an existing artifact at the target
+root or create it there. If the target is not writable, return the complete proposed artifact and
+state that it was not persisted.
+
+Before returning, confirm the artifact matches the requested mode, every resolved value has an
+origin, owner identifiers were inspected, exact measurements came from tools, invisible behavior
+was not passed from screenshots, gaps block application, and the manifest remains canonical runtime
+authority.

package/dist/tools/skills/assets/zora-designer/references/audit.md

@@ -0,0 +1,100 @@
+# Evidence-Aware Design Audit
+
+The canonical rubric, weights, criterion keys, status factors, confidence factors, and release-gate
+inventory live in [audit-rubric.json](../assets/audit-rubric.json). The calculation implementation
+in [audit.mjs](../scripts/audit.mjs) loads that file. Do not reproduce either list or the arithmetic
+in prompts, references, tests, or application code.
+
+Inspect the current catalog when needed:
+
+```text
+bun .agents/skills/zora-designer/scripts/audit.mjs catalog
+```
+
+## Evidence intake
+
+Every evidence item has a stable ID, kind, location or screen/viewport/state, precise observation,
+evidence level, confidence and factor, reproduction steps, and limitations. Preserve URL capture
+parameters and ordered image-series position.
+
+Use these evidence levels:
+
+- `measured`: reproducible source, computed runtime value, accessibility tree, instrumented test, or
+  reliable tool output;
+- `observed`: clearly visible behavior or high-resolution visual evidence;
+- `estimated`: approximate measurement with a known scale or compositing limitation;
+- `inferred`: plausible intent without direct proof;
+- `not-assessable`: the source cannot demonstrate the criterion.
+
+Confidence is explicit, not inferred from the evidence-level word. High-confidence computed runtime
+evidence and medium-confidence design metadata can both be measured in different senses.
+
+A visual-only audit may assess composition, hierarchy, apparent typography, approximate spacing,
+visible states, and color risk. It cannot positively verify semantic names, keyboard or focus order,
+screen-reader announcements, responsive reflow, text scaling, target dimensions without a trusted
+scale, loading feedback, recovery behavior, or performance.
+
+## Criterion assessment
+
+Assess all canonical criteria from the rubric asset. For each criterion record:
+
+- `applicable` and why;
+- `status` using the asset's canonical status keys;
+- `evidenceIds` and the non-empty `essentialEvidenceIds` needed to reproduce any assessed status;
+- `confidenceFactor` equal to the minimum factor among essential evidence;
+- `reason`.
+
+Unavailable evidence is `not-assessable`, not inapplicable. A criterion is inapplicable only when
+the product or audited scope genuinely has no such responsibility.
+
+Repeated instances influence severity and remediation but do not create repeated deductions. Map
+one root cause to one primary criterion and cross-reference related rules without scoring them
+again. C.R.A.P., visual dominance, implementation quality, performance, and user testing are useful
+cross-checks or delivery gates, not extra weighted rules.
+
+Run deterministic calculation through `calculateAudit` or the audit CLI. It derives rule totals,
+score, evidence coverage, possible range, aggregate confidence, release items, and the aggregate
+release gate from the canonical asset. Do not hand-calculate or round intermediate values.
+
+Interpretation rules:
+
+- A non-null score always appears with coverage, confidence, and possible range.
+- Low coverage makes the score provisional even when the score is high.
+- Confidence describes evidence quality; coverage describes how much applicable rubric weight was
+  assessed.
+- No applicable criteria yields null score, coverage, range, and confidence.
+- Applicable but wholly unassessed criteria yield null score, zero coverage, and a 0–100 range.
+- The numerical score never overrides a failed or unassessed release gate.
+
+## Release gates
+
+Populate every canonical release item and subcriterion from the asset. Missing evidence does not
+make a release subcriterion inapplicable. One item fails if any applicable subcriterion fails,
+passes only when at least one applies and all applicable subcriteria pass, and remains
+`not-assessable` when applicable evidence is incomplete. The aggregate gate fails on any item
+failure and passes only when all applicable items pass.
+
+Any confirmed applicable WCAG A/AA failure, keyboard trap, inaccessible core task, unrecoverable
+destructive flow, owner diagnostic blocker, or equivalent critical blocker fails the relevant gate
+even if the weighted score is high. This compact gate is not a formal WCAG conformance claim.
+
+## Findings, risks, and remediation
+
+A scored finding contains a stable ID, primary rule and criterion, severity, location, evidence,
+evidence level, expected result, user/task impact, deduplicated root cause, non-scoring related
+rules, smallest useful fix, verification, confidence, evidence IDs, and script-allocated score
+impact. Exact colors, ratios, selectors, and dimensions require measurement.
+
+Plausible but unconfirmed concerns are non-scoring risks. Each risk records its signal, evidence
+level, confidence, possible impact, and verification needed. It cannot lower the score or fail a
+release gate until evidence promotes it to a finding.
+
+Prioritize remediation in this order:
+
+1. release-gate, accessibility, safety, and core-task blockers;
+2. shared owner tokens or components that resolve multiple findings;
+3. major hierarchy, form, navigation, and recovery problems;
+4. localized usability and consistency issues;
+5. aesthetic polish.
+
+Improve the owning shared primitive before patching instances or inventing a Studio-only path.