xinetzone/SpecWeave / .agents/skills/solo-design
solo-design
VerifiedDesign website pages, UI screens, prototypes, page-level visual systems, and existing `.design` project mutations. Use for websites, landing pages, UI prototypes, redesigns, theme changes, 1:1 restoration, editable static layouts, and complex canvas projects. Route bitmap image edits to solo-image-edit, bitmap-first poster/banner/KV/cover generation to solo-graphic-generation, and reusable Design Library / Design System / token architecture creation to design-library-creator.
Solo Design
This Skill uses a decision -> load -> preflight -> dispatch -> validate -> readiness routing chain. SKILL.md only classifies the request, writes the resolved lane into working notes / summary when a project exists, and loads exactly one intent-workflows/<selected>/INTENT_WORKFLOW.md.
Do not scan all workflow folders. After a lane is selected, read only that lane's INTENT_WORKFLOW.md and the files declared in its Context Requirements. Before any dispatch, the selected lane's declared context files must be recorded with body-read evidence in runtime-orchestration-summary.json.project.contextRequirementsLoaded[].
Top-Level Skill Handoffs
Evaluate these handoffs in order before Route Priority and before reading any intent workflow.
Immediately invoke the Skill tool with name: "solo-image-edit" and stop applying this Skill when either condition matches:
- Current-turn context contains an image-region annotation whose parsed payload type is
design-image-annotation. Match this semantically across available envelopes such as aweb-elementrecord or aselected_browser_item; do not require one exact wrapper or field layout. - The user explicitly asks to edit, retouch, remove, replace, recolor, restyle, extend, or otherwise change pixels in an existing bitmap image, and an original image is available as a local file or current-turn attachment.
The structured design-image-annotation marker is authoritative even when the visible user text is only a short instruction such as "按评论改一下" or only names the selected image-region. Do not route ordinary HTML/DOM/CSS/page-component edits merely because the page contains images. Do not route text-to-image creation without an original bitmap; keep those requests in the graphic lanes below.
After the image-edit handoff does not match, immediately invoke the Skill tool with name: "solo-graphic-generation" and stop applying this Skill when the generated bitmap itself is the deliverable: poster, banner, KV/key visual, cover, invitation, social card, illustration, portrait, product image, transparent-background asset, cutout, or an image batch without editable HTML requirements.
The graphic-generation handoff also applies when the user asks to generate and append a new image asset to an existing canvas. solo-graphic-generation owns the required pre-generation size confirmation, image generation, canvas registration, and image-only validation. Keep text-critical static layouts in this Skill when exact readable/editable copy, DOM typography, tables, schedules, prices, legal copy, PPT, one-pager, or manual/document structure is required.
Global Invariants
- Output is canvas-first: every produced page or kept image asset must be represented in the
.designfile. .designvalidity is blocking: valid JSON, non-emptydata, existingdevMetadata.htmlSrc/imageSrc, unique node ids, and registered image assets.- Explicit user instructions win for design direction and optional process choices, but cannot override canvas validity, rendering correctness, safety, or required validation gates.
- Existing page/project changes create a new comparison page by default. In-place edit is allowed only when the user explicitly asks to overwrite, replace, or modify in place.
- New dispatches must never target
shared-runtime/blocked-legacy-entrypoints/blocked-legacy-page-template.md. shared-runtime/agent-dispatch-runtime/shared-page-rendering-kernel.mdis a Declared Load only; it is not an Always Load file.visual-experience/anddelivery-quality/are loaded only when the selectedINTENT_WORKFLOW.mddeclares them.- Flow-control rules are model-agnostic. Do not add model-specific branches, model names, or per-model instructions to this Skill.
- Production visual stability is a generation-time gate for layout-risk pages. Page Sub-Agents must build layout-critical structure with responsive-safe static CSS in the page, not rely only on Tailwind browser runtime compilation, late validation, or repair passes. If first-screen frame, density, overflow, or overlap checks cannot pass during generation, return
qualityGate: "failed|blocked"instead of producing a low-fidelity page for later repair.
Route Priority
After no top-level handoff matches, evaluate from P1 to P9 and stop at the first match.
| Priority | Condition | resolvedLane | Load |
|---|---|---|---|
| P1 | Existing project/page/selection edit, add state, add comparison page, navigation add, delete/adopt/refresh operation | existing_edit_add_comparison | intent-workflows/intent-project-mutation/INTENT_WORKFLOW.md |
| P2 | Redesign duplicate project or convert an existing project to another device/view | redesign_duplicate_project | intent-workflows/intent-project-mutation/INTENT_WORKFLOW.md |
| P3 | Generate variants, alternatives, multi-scheme comparison for existing pages | variants_multi_scheme | intent-workflows/intent-project-mutation/INTENT_WORKFLOW.md |
| P4 | Project-wide theme, token, color, typography, or head refresh customization | theme_customize | intent-workflows/intent-project-mutation/INTENT_WORKFLOW.md |
| P5 | Explicit 1:1, clone, replicate, pixel-perfect, exact restoration, or restore eval metadata | restore_1to1 | intent-workflows/intent-page-restore-1to1/INTENT_WORKFLOW.md |
| P6 | Active Design Library identity, Library path, componentPlan, or actualTokenNameReference | library_bound | intent-workflows/intent-page-library-bound/INTENT_WORKFLOW.md |
| P7 | graphicStrategyGate = "layout-static": long copy, PPT/one-pager/manual, exact editable text, table/schedule/legal/brand copy | graphic_layout_static | intent-workflows/intent-project-complex-build/INTENT_WORKFLOW.md |
| P8 | Long PRD, multi-page or multi-device project, generation tree, complex interaction/state expansion, fragment mode | complex_html_page | intent-workflows/intent-project-complex-build/INTENT_WORKFLOW.md |
| P9 | No Library, no restore target, open-ended page/UI/site/app creation | free_exploration | intent-workflows/intent-page-free-exploration/INTENT_WORKFLOW.md |
Conflict rule: earlier priorities win. Example: explicit restore wins over active Library; existing project mutation wins over new-project creation unless the user explicitly asks for a separate new project.
Dispatch Discipline
- Main Agent owns routing,
.designwrites, image-node registration, interaction registration, and validation. - Before any dispatch, Main Agent must record a Step 0 lane gate in
runtime-orchestration-summary.json.project:resolvedLane,selectedIntentWorkflowRead: true,contextRequirementsLoaded[],contextReadScope,validationRunDiscipline, andlowValueCallWatchdog. EachcontextRequirementsLoaded[]row must include{ path, readStatus: "loaded", bodyRead: true }. If this evidence cannot be written, do not dispatch. - Page Sub-Agents consume only the selected lane page runtime guide, the selected lane dispatch contract, and the packet fields provided by Main Agent.
- Page Sub-Agents do not write
.design, run project validators, start preview servers, create helper scripts, or generate images. - Page Sub-Agent packets must include
allowedWritePaths[]; completions must reportchangedFiles[]. The Main Agent owns.design,runtime-orchestration-summary.json, validation reports, readiness evidence, and final delivery text. - Cross-lane collaboration uses
runtime-orchestration-summary.json,dispatchPreflightManifest[], andsourceProjectLaneHeritage; do not make one intent workflow read another intent workflow's runbook. supplementaryReads[]entries must be explicit{ path, reason, ownerLane }. Free lane may keepimplementationEscalationReasononly as a compatibility alias.- Restore dispatches must lock
project.sourceAuthorityLockbefore page generation. Screenshot/user-provided imagery is visual authority when present; URL/browser evidence may supplement content but must not override the locked visual authority. - Static graphic / layout-heavy page dispatches must write
project.visualQualityCheckpoints[]before generation. These checkpoints lock visual anchor, information hierarchy, composition structure, and implementation strategy without adding open-ended exploration. - Layout-risk page dispatches must include a compact production stability target: target device/frame, viewport mode, first-screen content regions, responsive breakpoints, and high-risk compression/overlap risks. Page Sub-Agents must report
staticRenderingEvidenceandviewportIntegrityEvidence; missing evidence blocks that layout-risk dispatch before project validation. - Long reasoning without tool calls, artifact diffs, or structured decisions must follow
project.lowValueCallWatchdog.noProgressNextActionand move to readiness or a minimal blocked summary; do not continue thinking loops.
Allowed Scripts
Only Skill-provided deterministic scripts below are allowed during execution or maintenance:
| Script | Caller | When |
|---|---|---|
node {SKILL_DIR}/shared-runtime/deterministic-tooling/apply-html-head-contract.mjs | Page Sub-Agent / Main Agent fallback | Generate or replace approved HTML <head> |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/validate-restore-contract.mjs | Main Agent | Restore-only pre-dispatch contract gate; validates source identity, page state, measured facts, region groups, CSS aliases, and packet readiness |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/validate-design-workspace.mjs | Main Agent | Final validation for HTML/page projects |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/record-validation-repair-entry.mjs | Main Agent | Required first action after a failed validate-design-workspace.mjs report; records lightweight script-owned repair evidence. Do not hand-write hashes or validation history |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/record-dispatch-completion.mjs | Main Agent | Records Page Sub-Agent completion evidence into expectedDispatches[] and main-agent mutation evidence. Do not hand-edit dispatch completion JSON |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/validate-finish-readiness.mjs | Main Agent | Design artifact readiness gate after validation-report.json success=true; run --check=all for final delivery; checks validation report integrity, .design registration, page/image paths, interactions, post-validation mutation, and restore final-response text policy when provided |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/validate-design-file-format.mjs | Internal only | Called by workspace validation |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/validate-lane-runtime-contract.mjs | Main Agent / maintainer | Lane contract preflight and audit |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/build-page-dispatch-manifest.mjs | Main Agent | Build page dispatch preflight manifest for page-dispatch lanes |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/repair-mobile-navigation-flow.mjs | Main Agent | One-time repair after mobile navigation validation issue |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/check-intent-workflow-boundaries.mjs | Maintainer | Validate lane isolation and declared visual-quality loading |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/check-skill-runtime-read-scope.mjs | Maintainer | Validate agent-readable runbook file-size and read-scope limits |
node {SKILL_DIR}/shared-runtime/deterministic-tooling/check-solo-design-contract-regressions.mjs | Maintainer | Run fixture-based regressions for dispatch, repair ledger, version, and readiness contract failures |
Out-Of-Scope Routing
- Pixel-level edits of an existing bitmap, including structured Design image comments, go to
solo-image-edit. - Bitmap-first poster/banner/KV/cover/card/illustration/image-batch generation goes to
solo-graphic-generation. - Professional reusable Design Library / Design System / token architecture creation goes to
design-library-creator. - Legacy runtime templates are blocked. If any dispatch packet or task query names
blocked-legacy-page-template.md, stop and re-route to the lane-owned page runtime guide. create-design-systemstyle requests inside page creation are treated as page styling requirements only; they must not self-create a reusable Design Library.
Completion Gate
Before final delivery, run the validation required by the selected INTENT_WORKFLOW.md. For restore_1to1 workflows, skip validate-design-workspace.mjs entirely and proceed directly to validate-finish-readiness.mjs <designProjectPath> --check=all --final-response-file=<draft.md>; ensure the draft is link-free and path-free. For other HTML/page workflows, after validate-design-workspace.mjs --report-json=<designProjectPath>/validation-report.json succeeds, run validate-finish-readiness.mjs <designProjectPath> --check=all as the Design artifact readiness gate. Do not claim completion while validation has blocking errors, artifact readiness fails, validation report integrity fails, or restore final-response text policy fails.
If validate-design-workspace.mjs reports only repair-ledger metadata warnings, do not rerun full validation or inspect validator source. Repair-ledger metadata is diagnostic; continue only when validation-report.json success=true and validate-finish-readiness.mjs <designProjectPath> --check=all passes. If validation-report.json.terminalState is present, stop and report the blocking summary instead of starting another repair loop.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |