Update to VIXNODE
Snapshot Nov 9, 2026 · 06:07 UTC · version 3.5.0
Collection source: downloaded plugin package.
Instructions updated for vixnode-website-builder
Skill instructions — Now says: --- name: vixnode-website-builder description: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website...
Observed in instructions or declared skills. Runtime behavior has not been tested.
Up to 3 plugins free · no card required.
All fields and original values.
Skill instructions
Oct 24, 2026 · 12:56 UTC
--- name: vixnode-website-builder description: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, p...
Nov 9, 2026 · 06:07 UTC
--- name: vixnode-website-builder description: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, pro...
Supporting files
Oct 24, 2026 · 12:56 UTC
[{"relative_path":"agents/openai.yaml","size_in_bytes":519},{"relative_path":"references/art-direction.md","size_in_bytes":16880},{"relative_path":"references/page-creation.md","size_in_bytes":37699}]
Nov 9, 2026 · 06:07 UTC
[{"relative_path":"agents/openai.yaml","size_in_bytes":519},{"relative_path":"skills/art-direction.md","size_in_bytes":16880},{"relative_path":"skills/color-system.md","size_in_bytes":4699},{"relative_path":"skills/library/assets/gltf-as...
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -1,100 +1,114 @@ ---- -name: vixnode-website-builder -description: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website. ---- - -# VIXNODE Website Builder - -Use VIXNODE MCP tools to turn a website request into an approved, structurally valid result. The user's explicit requirements override defaults in this skill. - -Apply this workflow to website creation, modification, preview, and publication. Treat tool schemas and server validation as authoritative contracts: never bypass a required field, confirmation, page-type rule, layout relationship, unique path constraint, or preview token because the workflow text appears to permit it. - -## Load the Relevant Guidance - -- Before creating or changing VIXNODE page HTML, read [references/page-creation.md](references/page-creation.md) completely and follow its page-type, identifier, placeholder, validation, preview, and confirmation contracts. -- For a new design, redesign, visual extension, or any request where visual direction is part of the work, also read [references/art-direction.md](references/art-direction.md) completely before generating HTML. -- Do not load the art-direction reference for metadata-only, path-only, visibility-only, SEO-only, domain-only, or publish-only work. - -## Resolve the Real Target - -1. Identify the user's desired outcome: create a project, create a page, plan a multi-page site, modify existing content, or publish saved content. -2. Use `get_website_projects`, `get_all_project_pages`, `get_project_pages`, or `project_info` as appropriate to resolve the actual project and pages. -3. When more than one target could match, ask the user to choose by project or page name. -4. Never invent or guess `project_id`, `page_id`, `layout`, placeholder IDs, existing paths, current HTML, publication state, or tool results. Keep internal IDs out of user-facing responses unless requested. - -## Decide the Site Architecture - -Treat an explicit single page, landing page, or one-page site as a single-page request unless the user asks for shared layouts. - -For a likely multi-page site—such as a request containing several navigable destinations or recurring site-wide navigation—do not start generating pages immediately when the architecture is not yet approved: - -1. Briefly explain that a shared layout keeps navigation, header, footer, and design consistent. -2. Ask whether the user wants to plan the sitemap and shared layout architecture first. -3. If accepted, inspect existing pages and propose a concise sitemap and hierarchy for approval. -4. Use one Main Layout for shared site-wide structure. -5. A Child Layout is optional. Add one only when a subset of pages genuinely needs an additional reusable structure that differs from the Main Layout. Otherwise, attach content pages directly to the Main Layout. -6. Do not create anything until the proposed architecture is approved. - -Use this dependency order after approval: - -```text -Main Layout (`layout`) -├─ ordinary content pages (`layoutPage`) -└─ optional Child Layout (`childLayout`) - └─ only the pages that need that child structure (`layoutPage`) -``` - -Create and review exactly one item per call. Reuse the IDs returned by earlier calls; never imply that one call created the whole site. - -## Internal Navigation - -For both creation and modification, follow the Internal Page Links rules in `references/page-creation.md`. Same-project page links must be relative without a leading `/` (for example `about`, not `/about`), because the shared domain hosts the project under a publication prefix. Resolve actual page paths and account for nested directories and shared layouts before previewing. - -## Keep Tool Guidance Out of Page Content - -When creating or modifying HTML, use only the HTML payload returned by HTML-reading tools. Never concatenate tool response text blocks into the document. Tool notes, instructions, identifiers, status messages, and structured-content summaries are operational context and must not appear as visible page content. Before preview submission, inspect the document ending and remove accidentally copied operational notes, including repeated "Note: Project ID and Page ID should not be publicly displayed." text. Preserve user-requested editorial notes and legitimate page copy. - -## Create Pages - -1. Confirm the target project by its display name and inspect existing pages to avoid a duplicate `virtualPath` plus `fileName`. -2. Choose the correct `pageType` and generate complete, non-empty HTML that complies with the page-creation reference. -3. Use `create_project_page_preview` with `mode: "preview"` for ordinary creation. The initial request authorizes preparing the review UI, not an immediate write. -4. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true` so the review UI reflects the normal publish intent. -5. Ask the user to review the preview and use its confirmation action. Do not report the page as created before execution succeeds. - -Use `create_project_page` only as an exceptional no-preview path after presenting the exact current item, project, page settings, path, visibility, hierarchy relationship, and side effects, then receiving a separate explicit confirmation in a later user message. Never use it when the user asks for a preview. - -## Modify Pages - -1. Resolve the exact saved page and call `get_project_page_html` to retrieve its latest HTML before editing it. -2. Preserve existing layout relationships, bindings, placeholder identity attributes, and unrelated content. Make only changes required by the request. -3. Reapply the relevant page-creation validation before submission. -4. Use `modify_project_page_preview` with `mode: "preview"` for ordinary changes. -5. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true`. -6. Do not claim the update is applied until the confirmed execution succeeds. - -Use `modify_project_page` only after the same separate, explicit no-preview confirmation standard used for direct creation. A user's initial modification request is not that confirmation. - -## Visual Direction - -For design work, derive the visual system from the site's content, audience, brand, and purpose before generating HTML. Keep the visual language consistent across a multi-page site without making every page structurally identical. If the brand direction is materially ambiguous, ask only for the missing choices that would change the result; otherwise make a reasoned proposal and expose it for review. - -## Publish and Other Mutations - -- Use `publish_project_page` only when the user requests publishing a saved page or when publication is separately confirmed outside a preview workflow. -- Use `publish_project` only for intentional project-wide republication, such as propagating a changed layout to affected public pages. -- Use `validate_project_domain`, `configure_project_domain`, `get_project_domain_status`, `remove_project_domain`, `modify_project`, SEO tools, uploads, or media tools only when required by the user's website request. Explain material effects before mutation. -- Do not claim publication is complete until the tool confirms success. If it reports a pending state, report that state accurately. -- Do not retry a mutating failure with changed parameters unless the failure explains a safe correction or the user approves the revised action. - -## Completion Response - -Report only confirmed facts: - -- project and page names; -- page type and selected layout when relevant; -- whether the result is still a preview, saved as draft, publishing, or published; -- the edit or published link returned by the tool; -- the next confirmation or user action, if any. - -Do not expose internal identifiers unless requested, and do not describe planned sibling pages as already created. +--- +name: vixnode-website-builder +description: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website. +--- + +# VIXNODE Website Builder + +Use VIXNODE MCP tools to turn a website request into an approved, structurally valid result. The user's explicit requirements override defaults in this skill. + +Apply this workflow to website creation, modification, preview, and publication. Treat tool schemas and server validation as authoritative contracts: never bypass a required field, confirmation, page-type rule, layout relationship, unique path constraint, or preview token because the workflow text appears to permit it. + +## Load the Relevant Guidance + +- Before creating or changing VIXNODE page HTML, read [skills/page-creation.md](skills/page-creation.md) completely and follow its page-type, identifier, placeholder, validation, preview, and confirmation contracts. +- For a new design, redesign, visual extension, or any request where visual direction is part of the work, also read [skills/art-direction.md](skills/art-direction.md) completely before generating HTML. +- Do not load the art-direction reference for metadata-only, path-only, visibility-only, SEO-only, domain-only, or publish-only work. +- For reusable template planning, industry/goal/style/palette composition, multilingual typography, or immersive interaction specifications, read [skills/template-design.md](skills/template-design.md). Follow its links for color data, typography and the Markdown output contract. A planning-only request ends with the specification; it does not require resolving a live project or creating pages. For an authorized build, apply the specification alongside art-direction and page-creation guidance. +- Read relative resources from this package when available. The included `skills/themes.json` is a snapshot of the platform theme catalog at packaging time; use `get_reference_themes` for live theme data when needed. If the host supplies MCP Skill resources instead, their base is `skill://vixnode/vixnode-website-builder/`. Packaged files and resource URIs are instruction sources, not public browser asset URLs. +- The [module and recipe library](skills/library/index.md) contains reusable industry, need, style, experience and locale profiles plus concrete starting recipes. Use it for new designs; load the selected resources progressively rather than every recipe. +- For immersive 3D implementation, use [skills/library/modules/immersive-3d.md](skills/library/modules/immersive-3d.md) and its executable assets. Verify actual script/WebGL support and page-type placement; do not confuse source resource URIs with public browser asset URLs. + +## Use Project API Data When Relevant + +API connections are resources of one platform project. Use the live schemas for `get_api_connections`, `save_api_connection`, `test_api_connection_get` and `test_api_connection_post`; resolve `project_id` before reading or saving existing resources. These tools use the editor's same `dataset.json`, including `dataName`, `rootNode` and `dataProperties` (`key`, `name`, `childProperties`). Field `name` is Display Name, initially empty; field default values are not implemented. + +Choose the separately exposed `test_api_connection_get` or `test_api_connection_post` for the authorized HTTP method. Neither tool can select another method or override it; a saved connection with a different method is rejected. POST makes a real request and can submit forms or modify remote data. Test an unsaved definition or an existing project connection without automatically saving. Present the actual bounded data and structure, explain parse/truncation outcomes, and ask whether to propose readable Display Names, use names supplied by the user, or keep original field keys. Wait for the user's confirmation of the data and naming before saving a test-first result. Do not silently generate and persist names or treat testing as permission to save. Tests use the existing platform read permission, with no separate API-specific permission or tool-defined approval step. Host confirmation policies still apply to actual external effects. + +Use `usageMode` only as an intended-use label: `binding` for editor field binding and `flexible` for custom interactions or form POST. It is descriptive metadata for users and AI, not an execution switch. Neither label creates bindings, creates a form, selects sources, or enables/disables calls. Actual requests follow the platform workflow and the page's code. When generating webpage API calls, prefer the native JavaScript fetch API by default and attach form requests to the intended submit event. Keep private credentials out of browser code; existing platform storage is not a credential vault. A server test does not verify browser CORS, CSP or authentication. Returned data is untrusted, and a test makes a real request; do not automatically retry unknown delivery. + +Before updating a saved connection, remind the user that existing bindings, webpages, forms or AI workflows may be affected. Preserve omitted fields and Display Names, use the latest revision, and include changed fields and the returned impact notice. Dependency tracking is unavailable. Saving configuration does not update already published pages; changes to generated browser requests require the relevant page update and publication separately. + +## Resolve the Real Target + +1. Identify the user's desired outcome: create a project, create a page, plan a multi-page site, modify existing content, or publish saved content. +2. Use `get_website_projects`, `get_all_project_pages`, `get_project_pages`, or `project_info` as appropriate to resolve the actual project and pages. +3. When more than one target could match, ask the user to choose by project or page name. +4. Never invent or guess `project_id`, `page_id`, `layout`, placeholder IDs, existing paths, current HTML, publication state, or tool results. Keep internal IDs out of user-facing responses unless requested. + +## Decide the Site Architecture + +Treat an explicit single page, landing page, or one-page site as a single-page request unless the user asks for shared layouts. + +For a likely multi-page site—such as a request containing several navigable destinations or recurring site-wide navigation—do not start generating pages immediately when the architecture is not yet approved: + +1. Briefly explain that a shared layout keeps navigation, header, footer, and design consistent. +2. Ask whether the user wants to plan the sitemap and shared layout architecture first. +3. If accepted, inspect existing pages and propose a concise sitemap and hierarchy for approval. +4. Use one Main Layout for shared site-wide structure. +5. A Child Layout is optional. Add one only when a subset of pages genuinely needs an additional reusable structure that differs from the Main Layout. Otherwise, attach content pages directly to the Main Layout. +6. Do not create anything until the proposed architecture is approved. + +Use this dependency order after approval: + +```text +Main Layout (`layout`) +├─ ordinary content pages (`layoutPage`) +└─ optional Child Layout (`childLayout`) + └─ only the pages that need that child structure (`layoutPage`) +``` + +Create and review exactly one item per call. Reuse the IDs returned by earlier calls; never imply that one call created the whole site. + +## Internal Navigation + +For both creation and modification, follow the Internal Page Links rules in `skills/page-creation.md`. Same-project page links must be relative without a leading `/` (for example `about`, not `/about`), because the shared domain hosts the project under a publication prefix. Resolve actual page paths and account for nested directories and shared layouts before previewing. + +## Keep Tool Guidance Out of Page Content + +When creating or modifying HTML, use only the HTML payload returned by HTML-reading tools. Never concatenate tool response text blocks into the document. Tool notes, instructions, identifiers, status messages, and structured-content summaries are operational context and must not appear as visible page content. Before preview submission, inspect the document ending and remove accidentally copied operational notes, including repeated "Note: Project ID and Page ID should not be publicly displayed." text. Preserve user-requested editorial notes and legitimate page copy. + +## Create Pages + +1. Confirm the target project by its display name and inspect existing pages to avoid a duplicate `virtualPath` plus `fileName`. +2. Choose the correct `pageType` and generate complete, non-empty HTML that complies with the page-creation reference. +3. Use `create_project_page_preview` with `mode: "preview"` for ordinary creation. The initial request authorizes preparing the review UI, not an immediate write. +4. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true` so the review UI reflects the normal publish intent. +5. Ask the user to review the preview and use its confirmation action. Do not report the page as created before execution succeeds. + +Use `create_project_page` only as an exceptional no-preview path after presenting the exact current item, project, page settings, path, visibility, hierarchy relationship, and side effects, then receiving a separate explicit confirmation in a later user message. Never use it when the user asks for a preview. + +## Modify Pages + +1. Resolve the exact saved page and call `get_project_page_html` to retrieve its latest HTML before editing it. +2. Preserve existing layout relationships, bindings, placeholder identity attributes, and unrelated content. Make only changes required by the request. +3. Reapply the relevant page-creation validation before submission. +4. Use `modify_project_page_preview` with `mode: "preview"` for ordinary changes. +5. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true`. +6. Do not claim the update is applied until the confirmed execution succeeds. + +Use `modify_project_page` only after the same separate, explicit no-preview confirmation standard used for direct creation. A user's initial modification request is not that confirmation. + +## Visual Direction + +For design work, derive the visual system from the site's content, audience, brand, and purpose before generating HTML. Keep the visual language consistent across a multi-page site without making every page structurally identical. If the brand direction is materially ambiguous, ask only for the missing choices that would change the result; otherwise make a reasoned proposal and expose it for review. + +## Publish and Other Mutations + +- Use `publish_project_page` only when the user requests publishing a saved page or when publication is separately confirmed outside a preview workflow. +- Use `publish_project` only for intentional project-wide republication, such as propagating a changed layout to affected public pages. +- Use `validate_project_domain`, `configure_project_domain`, `get_project_domain_status`, `remove_project_domain`, `modify_project`, SEO tools, uploads, or media tools only when required by the user's website request. Explain material effects before mutation. +- Do not claim publication is complete until the tool confirms success. If it reports a pending state, report that state accurately. +- Do not retry a mutating failure with changed parameters unless the failure explains a safe correction or the user approves the revised action. + +## Completion Response + +Report only confirmed facts: + +- project and page names; +- page type and selected layout when relevant; +- whether the result is still a preview, saved as draft, publishing, or published; +- the edit or published link returned by the tool; +- the next confirmation or user action, if any. + +Do not expose internal identifiers unless requested, and do not describe planned sibling pages as already created.
Full technical diff · 2 changed fields
changed /included_files
[
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 519
},
{
"relative_path": "references/art-direction.md",
"size_in_bytes": 16880
},
{
"relative_path": "references/page-creation.md",
"size_in_bytes": 37699
}
][
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 519
},
{
"relative_path": "skills/art-direction.md",
"size_in_bytes": 16880
},
{
"relative_path": "skills/color-system.md",
"size_in_bytes": 4699
},
{
"relative_path": "skills/library/assets/gltf-asset.js",
"size_in_bytes": 3751
},
{
"relative_path": "skills/library/assets/immersive-demo.html",
"size_in_bytes": 4605
},
{
"relative_path": "skills/library/assets/immersive-math.js",
"size_in_bytes": 3171
},
{
"relative_path": "skills/library/assets/immersive-runtime.js",
"size_in_bytes": 9078
},
{
"relative_path": "skills/library/assets/immersive-scene.js",
"size_in_bytes": 6190
},
{
"relative_path": "skills/library/assets/immersive.css",
"size_in_bytes": 5916
},
{
"relative_path": "skills/library/composition.md",
"size_in_bytes": 5752
},
{
"relative_path": "skills/library/index.md",
"size_in_bytes": 5987
},
{
"relative_path": "skills/library/modules/asset-performance.md",
"size_in_bytes": 4826
},
{
"relative_path": "skills/library/modules/experiences.md",
"size_in_bytes": 5722
},
{
"relative_path": "skills/library/modules/immersive-3d.md",
"size_in_bytes": 5806
},
{
"relative_path": "skills/library/modules/immersive-integration.md",
"size_in_bytes": 4866
},
{
"relative_path": "skills/library/modules/industries.md",
"size_in_bytes": 6812
},
{
"relative_path": "skills/library/modules/locale-profiles.md",
"size_in_bytes": 4419
},
{
"relative_path": "skills/library/modules/needs.md",
"size_in_bytes": 5442
},
{
"relative_path": "skills/library/modules/scene-contract.md",
"size_in_bytes": 5292
},
{
"relative_path": "skills/library/modules/spatial-composition.md",
"size_in_bytes": 6083
},
{
"relative_path": "skills/library/modules/visual-styles.md",
"size_in_bytes": 8627
},
{
"relative_path": "skills/library/recipes/assembly-explainer.md",
"size_in_bytes": 3691
},
{
"relative_path": "skills/library/recipes/calm-consultation.md",
"size_in_bytes": 8313
},
{
"relative_path": "skills/library/recipes/collection-studio.md",
"size_in_bytes": 8186
},
{
"relative_path": "skills/library/recipes/craft-fieldnotes.md",
"size_in_bytes": 8102
},
{
"relative_path": "skills/library/recipes/cultural-passage.md",
"size_in_bytes": 8308
},
{
"relative_path": "skills/library/recipes/dining-atelier.md",
"size_in_bytes": 8024
},
{
"relative_path": "skills/library/recipes/industrial-proof.md",
"size_in_bytes": 8045
},
{
"relative_path": "skills/library/recipes/learning-pathways.md",
"size_in_bytes": 8270
},
{
"relative_path": "skills/library/recipes/object-study.md",
"size_in_bytes": 4134
},
{
"relative_path": "skills/library/recipes/product-workbench.md",
"size_in_bytes": 8263
},
{
"relative_path": "skills/library/recipes/public-impact.md",
"size_in_bytes": 8206
},
{
"relative_path": "skills/library/recipes/spatial-archive.md",
"size_in_bytes": 7918
},
{
"relative_path": "skills/library/recipes/spatial-passage.md",
"size_in_bytes": 4046
},
{
"relative_path": "skills/library/recipes/spatial-weave.md",
"size_in_bytes": 5083
},
{
"relative_path": "skills/page-creation.md",
"size_in_bytes": 36853
},
{
"relative_path": "skills/template-design.md",
"size_in_bytes": 7428
},
{
"relative_path": "skills/template-specification.md",
"size_in_bytes": 5540
},
{
"relative_path": "skills/template.md",
"size_in_bytes": 4198
},
{
"relative_path": "skills/themes.json",
"size_in_bytes": 20605
},
{
"relative_path": "skills/typography.md",
"size_in_bytes": 3344
}
]changed /skill_md_contents
"---\r\nname: vixnode-website-builder\r\ndescription: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website.\r\n---\r\n\r\n# VIXNODE Website Builder\r\n\r\nUse VIXNODE MCP tools to turn a website request into an approved, structurally valid result. The user's explicit requirements override defaults in this skill.\r\n\r\nApply this workflow to website creation, modification, preview, and publication. Treat tool schemas and server validation as authoritative contracts: never bypass a required field, confirmation, page-type rule, layout relationship, unique path constraint, or preview token because the workflow text appears to permit it.\r\n\r\n## Load the Relevant Guidance\r\n\r\n- Before creating or changing VIXNODE page HTML, read [references/page-creation.md](references/page-creation.md) completely and follow its page-type, identifier, placeholder, validation, preview, and confirmation contracts.\r\n- For a new design, redesign, visual extension, or any request where visual direction is part of the work, also read [references/art-direction.md](references/art-direction.md) completely before generating HTML.\r\n- Do not load the art-direction reference for metadata-only, path-only, visibility-only, SEO-only, domain-only, or publish-only work.\r\n\r\n## Resolve the Real Target\r\n\r\n1. Identify the user's desired outcome: create a project, create a page, plan a multi-page site, modify existing content, or publish saved content.\r\n2. Use `get_website_projects`, `get_all_project_pages`, `get_project_pages`, or `project_info` as appropriate to resolve the actual project and pages.\r\n3. When more than one target could match, ask the user to choose by project or page name.\r\n4. Never invent or guess `project_id`, `page_id`, `layout`, placeholder IDs, existing paths, current HTML, publication state, or tool results. Keep internal IDs out of user-facing responses unless requested.\r\n\r\n## Decide the Site Architecture\r\n\r\nTreat an explicit single page, landing page, or one-page site as a single-page request unless the user asks for shared layouts.\r\n\r\nFor a likely multi-page site—such as a request containing several navigable destinations or recurring site-wide navigation—do not start generating pages immediately when the architecture is not yet approved:\r\n\r\n1. Briefly explain that a shared layout keeps navigation, header, footer, and design consistent.\r\n2. Ask whether the user wants to plan the sitemap and shared layout architecture first.\r\n3. If accepted, inspect existing pages and propose a concise sitemap and hierarchy for approval.\r\n4. Use one Main Layout for shared site-wide structure.\r\n5. A Child Layout is optional. Add one only when a subset of pages genuinely needs an additional reusable structure that differs from the Main Layout. Otherwise, attach content pages directly to the Main Layout.\r\n6. Do not create anything until the proposed architecture is approved.\r\n\r\nUse this dependency order after approval:\r\n\r\n```text\r\nMain Layout (`layout`)\r\n├─ ordinary content pages (`layoutPage`)\r\n└─ optional Child Layout (`childLayout`)\r\n └─ only the pages that need that child structure (`layoutPage`)\r\n```\r\n\r\nCreate and review exactly one item per call. Reuse the IDs returned by earlier calls; never imply that one call created the whole site.\r\n\r\n## Internal Navigation\r\n\r\nFor both creation and modification, follow the Internal Page Links rules in `references/page-creation.md`. Same-project page links must be relative without a leading `/` (for example `about`, not `/about`), because the shared domain hosts the project under a publication prefix. Resolve actual page paths and account for nested directories and shared layouts before previewing.\r\n\r\n## Keep Tool Guidance Out of Page Content\r\n\r\nWhen creating or modifying HTML, use only the HTML payload returned by HTML-reading tools. Never concatenate tool response text blocks into the document. Tool notes, instructions, identifiers, status messages, and structured-content summaries are operational context and must not appear as visible page content. Before preview submission, inspect the document ending and remove accidentally copied operational notes, including repeated \"Note: Project ID and Page ID should not be publicly displayed.\" text. Preserve user-requested editorial notes and legitimate page copy.\r\n\r\n## Create Pages\r\n\r\n1. Confirm the target project by its display name and inspect existing pages to avoid a duplicate `virtualPath` plus `fileName`.\r\n2. Choose the correct `pageType` and generate complete, non-empty HTML that complies with the page-creation reference.\r\n3. Use `create_project_page_preview` with `mode: \"preview\"` for ordinary creation. The initial request authorizes preparing the review UI, not an immediate write.\r\n4. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true` so the review UI reflects the normal publish intent.\r\n5. Ask the user to review the preview and use its confirmation action. Do not report the page as created before execution succeeds.\r\n\r\nUse `create_project_page` only as an exceptional no-preview path after presenting the exact current item, project, page settings, path, visibility, hierarchy relationship, and side effects, then receiving a separate explicit confirmation in a later user message. Never use it when the user asks for a preview.\r\n\r\n## Modify Pages\r\n\r\n1. Resolve the exact saved page and call `get_project_page_html` to retrieve its latest HTML before editing it.\r\n2. Preserve existing layout relationships, bindings, placeholder identity attributes, and unrelated content. Make only changes required by the request.\r\n3. Reapply the relevant page-creation validation before submission.\r\n4. Use `modify_project_page_preview` with `mode: \"preview\"` for ordinary changes.\r\n5. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true`.\r\n6. Do not claim the update is applied until the confirmed execution succeeds.\r\n\r\nUse `modify_project_page` only after the same separate, explicit no-preview confirmation standard used for direct creation. A user's initial modification request is not that confirmation.\r\n\r\n## Visual Direction\r\n\r\nFor design work, derive the visual system from the site's content, audience, brand, and purpose before generating HTML. Keep the visual language consistent across a multi-page site without making every page structurally identical. If the brand direction is materially ambiguous, ask only for the missing choices that would change the result; otherwise make a reasoned proposal and expose it for review.\r\n\r\n## Publish and Other Mutations\r\n\r\n- Use `publish_project_page` only when the user requests publishing a saved page or when publication is separately confirmed outside a preview workflow.\r\n- Use `publish_project` only for intentional project-wide republication, such as propagating a changed layout to affected public pages.\r\n- Use `validate_project_domain`, `configure_project_domain`, `get_project_domain_status`, `remove_project_domain`, `modify_project`, SEO tools, uploads, or media tools only when required by the user's website request. Explain material effects before mutation.\r\n- Do not claim publication is complete until the tool confirms success. If it reports a pending state, report that state accurately.\r\n- Do not retry a mutating failure with changed parameters unless the failure explains a safe correction or the user approves the revised action.\r\n\r\n## Completion Response\r\n\r\nReport only confirmed facts:\r\n\r\n- project and page names;\r\n- page type and selected layout when relevant;\r\n- whether the result is still a preview, saved as draft, publishing, or published;\r\n- the edit or published link returned by the tool;\r\n- the next confirmation or user action, if any.\r\n\r\nDo not expose internal identifiers unless requested, and do not describe planned sibling pages as already created.\r\n"
"---\nname: vixnode-website-builder\ndescription: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website.\n---\n\n# VIXNODE Website Builder\n\nUse VIXNODE MCP tools to turn a website request into an approved, structurally valid result. The user's explicit requirements override defaults in this skill.\n\nApply this workflow to website creation, modification, preview, and publication. Treat tool schemas and server validation as authoritative contracts: never bypass a required field, confirmation, page-type rule, layout relationship, unique path constraint, or preview token because the workflow text appears to permit it.\n\n## Load the Relevant Guidance\n\n- Before creating or changing VIXNODE page HTML, read [skills/page-creation.md](skills/page-creation.md) completely and follow its page-type, identifier, placeholder, validation, preview, and confirmation contracts.\n- For a new design, redesign, visual extension, or any request where visual direction is part of the work, also read [skills/art-direction.md](skills/art-direction.md) completely before generating HTML.\n- Do not load the art-direction reference for metadata-only, path-only, visibility-only, SEO-only, domain-only, or publish-only work.\n- For reusable template planning, industry/goal/style/palette composition, multilingual typography, or immersive interaction specifications, read [skills/template-design.md](skills/template-design.md). Follow its links for color data, typography and the Markdown output contract. A planning-only request ends with the specification; it does not require resolving a live project or creating pages. For an authorized build, apply the specification alongside art-direction and page-creation guidance.\n- Read relative resources from this package when available. The included `skills/themes.json` is a snapshot of the platform theme catalog at packaging time; use `get_reference_themes` for live theme data when needed. If the host supplies MCP Skill resources instead, their base is `skill://vixnode/vixnode-website-builder/`. Packaged files and resource URIs are instruction sources, not public browser asset URLs.\n- The [module and recipe library](skills/library/index.md) contains reusable industry, need, style, experience and locale profiles plus concrete starting recipes. Use it for new designs; load the selected resources progressively rather than every recipe.\n- For immersive 3D implementation, use [skills/library/modules/immersive-3d.md](skills/library/modules/immersive-3d.md) and its executable assets. Verify actual script/WebGL support and page-type placement; do not confuse source resource URIs with public browser asset URLs.\n\n## Use Project API Data When Relevant\n\nAPI connections are resources of one platform project. Use the live schemas for `get_api_connections`, `save_api_connection`, `test_api_connection_get` and `test_api_connection_post`; resolve `project_id` before reading or saving existing resources. These tools use the editor's same `dataset.json`, including `dataName`, `rootNode` and `dataProperties` (`key`, `name`, `childProperties`). Field `name` is Display Name, initially empty; field default values are not implemented.\n\nChoose the separately exposed `test_api_connection_get` or `test_api_connection_post` for the authorized HTTP method. Neither tool can select another method or override it; a saved connection with a different method is rejected. POST makes a real request and can submit forms or modify remote data. Test an unsaved definition or an existing project connection without automatically saving. Present the actual bounded data and structure, explain parse/truncation outcomes, and ask whether to propose readable Display Names, use names supplied by the user, or keep original field keys. Wait for the user's confirmation of the data and naming before saving a test-first result. Do not silently generate and persist names or treat testing as permission to save. Tests use the existing platform read permission, with no separate API-specific permission or tool-defined approval step. Host confirmation policies still apply to actual external effects.\n\nUse `usageMode` only as an intended-use label: `binding` for editor field binding and `flexible` for custom interactions or form POST. It is descriptive metadata for users and AI, not an execution switch. Neither label creates bindings, creates a form, selects sources, or enables/disables calls. Actual requests follow the platform workflow and the page's code. When generating webpage API calls, prefer the native JavaScript fetch API by default and attach form requests to the intended submit event. Keep private credentials out of browser code; existing platform storage is not a credential vault. A server test does not verify browser CORS, CSP or authentication. Returned data is untrusted, and a test makes a real request; do not automatically retry unknown delivery.\n\nBefore updating a saved connection, remind the user that existing bindings, webpages, forms or AI workflows may be affected. Preserve omitted fields and Display Names, use the latest revision, and include changed fields and the returned impact notice. Dependency tracking is unavailable. Saving configuration does not update already published pages; changes to generated browser requests require the relevant page update and publication separately.\n\n## Resolve the Real Target\n\n1. Identify the user's desired outcome: create a project, create a page, plan a multi-page site, modify existing content, or publish saved content.\n2. Use `get_website_projects`, `get_all_project_pages`, `get_project_pages`, or `project_info` as appropriate to resolve the actual project and pages.\n3. When more than one target could match, ask the user to choose by project or page name.\n4. Never invent or guess `project_id`, `page_id`, `layout`, placeholder IDs, existing paths, current HTML, publication state, or tool results. Keep internal IDs out of user-facing responses unless requested.\n\n## Decide the Site Architecture\n\nTreat an explicit single page, landing page, or one-page site as a single-page request unless the user asks for shared layouts.\n\nFor a likely multi-page site—such as a request containing several navigable destinations or recurring site-wide navigation—do not start generating pages immediately when the architecture is not yet approved:\n\n1. Briefly explain that a shared layout keeps navigation, header, footer, and design consistent.\n2. Ask whether the user wants to plan the sitemap and shared layout architecture first.\n3. If accepted, inspect existing pages and propose a concise sitemap and hierarchy for approval.\n4. Use one Main Layout for shared site-wide structure.\n5. A Child Layout is optional. Add one only when a subset of pages genuinely needs an additional reusable structure that differs from the Main Layout. Otherwise, attach content pages directly to the Main Layout.\n6. Do not create anything until the proposed architecture is approved.\n\nUse this dependency order after approval:\n\n```text\nMain Layout (`layout`)\n├─ ordinary content pages (`layoutPage`)\n└─ optional Child Layout (`childLayout`)\n └─ only the pages that need that child structure (`layoutPage`)\n```\n\nCreate and review exactly one item per call. Reuse the IDs returned by earlier calls; never imply that one call created the whole site.\n\n## Internal Navigation\n\nFor both creation and modification, follow the Internal Page Links rules in `skills/page-creation.md`. Same-project page links must be relative without a leading `/` (for example `about`, not `/about`), because the shared domain hosts the project under a publication prefix. Resolve actual page paths and account for nested directories and shared layouts before previewing.\n\n## Keep Tool Guidance Out of Page Content\n\nWhen creating or modifying HTML, use only the HTML payload returned by HTML-reading tools. Never concatenate tool response text blocks into the document. Tool notes, instructions, identifiers, status messages, and structured-content summaries are operational context and must not appear as visible page content. Before preview submission, inspect the document ending and remove accidentally copied operational notes, including repeated \"Note: Project ID and Page ID should not be publicly displayed.\" text. Preserve user-requested editorial notes and legitimate page copy.\n\n## Create Pages\n\n1. Confirm the target project by its display name and inspect existing pages to avoid a duplicate `virtualPath` plus `fileName`.\n2. Choose the correct `pageType` and generate complete, non-empty HTML that complies with the page-creation reference.\n3. Use `create_project_page_preview` with `mode: \"preview\"` for ordinary creation. The initial request authorizes preparing the review UI, not an immediate write.\n4. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true` so the review UI reflects the normal publish intent.\n5. Ask the user to review the preview and use its confirmation action. Do not report the page as created before execution succeeds.\n\nUse `create_project_page` only as an exceptional no-preview path after presenting the exact current item, project, page settings, path, visibility, hierarchy relationship, and side effects, then receiving a separate explicit confirmation in a later user message. Never use it when the user asks for a preview.\n\n## Modify Pages\n\n1. Resolve the exact saved page and call `get_project_page_html` to retrieve its latest HTML before editing it.\n2. Preserve existing layout relationships, bindings, placeholder identity attributes, and unrelated content. Make only changes required by the request.\n3. Reapply the relevant page-creation validation before submission.\n4. Use `modify_project_page_preview` with `mode: \"preview\"` for ordinary changes.\n5. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true`.\n6. Do not claim the update is applied until the confirmed execution succeeds.\n\nUse `modify_project_page` only after the same separate, explicit no-preview confirmation standard used for direct creation. A user's initial modification request is not that confirmation.\n\n## Visual Direction\n\nFor design work, derive the visual system from the site's content, audience, brand, and purpose before generating HTML. Keep the visual language consistent across a multi-page site without making every page structurally identical. If the brand direction is materially ambiguous, ask only for the missing choices that would change the result; otherwise make a reasoned proposal and expose it for review.\n\n## Publish and Other Mutations\n\n- Use `publish_project_page` only when the user requests publishing a saved page or when publication is separately confirmed outside a preview workflow.\n- Use `publish_project` only for intentional project-wide republication, such as propagating a changed layout to affected public pages.\n- Use `validate_project_domain`, `configure_project_domain`, `get_project_domain_status`, `remove_project_domain`, `modify_project`, SEO tools, uploads, or media tools only when required by the user's website request. Explain material effects before mutation.\n- Do not claim publication is complete until the tool confirms success. If it reports a pending state, report that state accurately.\n- Do not retry a mutating failure with changed parameters unless the failure explains a safe correction or the user approves the revised action.\n\n## Completion Response\n\nReport only confirmed facts:\n\n- project and page names;\n- page type and selected layout when relevant;\n- whether the result is still a preview, saved as draft, publishing, or published;\n- the edit or published link returned by the tool;\n- the next confirmation or user action, if any.\n\nDo not expose internal identifiers unless requested, and do not describe planned sibling pages as already created.\n"
Full snapshot data
{
"description": "Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website.",
"included_files": [
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 519
},
{
"relative_path": "skills/art-direction.md",
"size_in_bytes": 16880
},
{
"relative_path": "skills/color-system.md",
"size_in_bytes": 4699
},
{
"relative_path": "skills/library/assets/gltf-asset.js",
"size_in_bytes": 3751
},
{
"relative_path": "skills/library/assets/immersive-demo.html",
"size_in_bytes": 4605
},
{
"relative_path": "skills/library/assets/immersive-math.js",
"size_in_bytes": 3171
},
{
"relative_path": "skills/library/assets/immersive-runtime.js",
"size_in_bytes": 9078
},
{
"relative_path": "skills/library/assets/immersive-scene.js",
"size_in_bytes": 6190
},
{
"relative_path": "skills/library/assets/immersive.css",
"size_in_bytes": 5916
},
{
"relative_path": "skills/library/composition.md",
"size_in_bytes": 5752
},
{
"relative_path": "skills/library/index.md",
"size_in_bytes": 5987
},
{
"relative_path": "skills/library/modules/asset-performance.md",
"size_in_bytes": 4826
},
{
"relative_path": "skills/library/modules/experiences.md",
"size_in_bytes": 5722
},
{
"relative_path": "skills/library/modules/immersive-3d.md",
"size_in_bytes": 5806
},
{
"relative_path": "skills/library/modules/immersive-integration.md",
"size_in_bytes": 4866
},
{
"relative_path": "skills/library/modules/industries.md",
"size_in_bytes": 6812
},
{
"relative_path": "skills/library/modules/locale-profiles.md",
"size_in_bytes": 4419
},
{
"relative_path": "skills/library/modules/needs.md",
"size_in_bytes": 5442
},
{
"relative_path": "skills/library/modules/scene-contract.md",
"size_in_bytes": 5292
},
{
"relative_path": "skills/library/modules/spatial-composition.md",
"size_in_bytes": 6083
},
{
"relative_path": "skills/library/modules/visual-styles.md",
"size_in_bytes": 8627
},
{
"relative_path": "skills/library/recipes/assembly-explainer.md",
"size_in_bytes": 3691
},
{
"relative_path": "skills/library/recipes/calm-consultation.md",
"size_in_bytes": 8313
},
{
"relative_path": "skills/library/recipes/collection-studio.md",
"size_in_bytes": 8186
},
{
"relative_path": "skills/library/recipes/craft-fieldnotes.md",
"size_in_bytes": 8102
},
{
"relative_path": "skills/library/recipes/cultural-passage.md",
"size_in_bytes": 8308
},
{
"relative_path": "skills/library/recipes/dining-atelier.md",
"size_in_bytes": 8024
},
{
"relative_path": "skills/library/recipes/industrial-proof.md",
"size_in_bytes": 8045
},
{
"relative_path": "skills/library/recipes/learning-pathways.md",
"size_in_bytes": 8270
},
{
"relative_path": "skills/library/recipes/object-study.md",
"size_in_bytes": 4134
},
{
"relative_path": "skills/library/recipes/product-workbench.md",
"size_in_bytes": 8263
},
{
"relative_path": "skills/library/recipes/public-impact.md",
"size_in_bytes": 8206
},
{
"relative_path": "skills/library/recipes/spatial-archive.md",
"size_in_bytes": 7918
},
{
"relative_path": "skills/library/recipes/spatial-passage.md",
"size_in_bytes": 4046
},
{
"relative_path": "skills/library/recipes/spatial-weave.md",
"size_in_bytes": 5083
},
{
"relative_path": "skills/page-creation.md",
"size_in_bytes": 36853
},
{
"relative_path": "skills/template-design.md",
"size_in_bytes": 7428
},
{
"relative_path": "skills/template-specification.md",
"size_in_bytes": 5540
},
{
"relative_path": "skills/template.md",
"size_in_bytes": 4198
},
{
"relative_path": "skills/themes.json",
"size_in_bytes": 20605
},
{
"relative_path": "skills/typography.md",
"size_in_bytes": 3344
}
],
"name": "vixnode-website-builder",
"skill_md_contents": "---\nname: vixnode-website-builder\ndescription: Plan, design, create, modify, preview, and publish single-page or multi-page VIXNODE websites. Use for website architecture, main and optional child layouts, page HTML, visual direction, project-page editing, preview approval, and publication through VIXNODE. Do not use for account, quota, or media-library requests that do not involve a website.\n---\n\n# VIXNODE Website Builder\n\nUse VIXNODE MCP tools to turn a website request into an approved, structurally valid result. The user's explicit requirements override defaults in this skill.\n\nApply this workflow to website creation, modification, preview, and publication. Treat tool schemas and server validation as authoritative contracts: never bypass a required field, confirmation, page-type rule, layout relationship, unique path constraint, or preview token because the workflow text appears to permit it.\n\n## Load the Relevant Guidance\n\n- Before creating or changing VIXNODE page HTML, read [skills/page-creation.md](skills/page-creation.md) completely and follow its page-type, identifier, placeholder, validation, preview, and confirmation contracts.\n- For a new design, redesign, visual extension, or any request where visual direction is part of the work, also read [skills/art-direction.md](skills/art-direction.md) completely before generating HTML.\n- Do not load the art-direction reference for metadata-only, path-only, visibility-only, SEO-only, domain-only, or publish-only work.\n- For reusable template planning, industry/goal/style/palette composition, multilingual typography, or immersive interaction specifications, read [skills/template-design.md](skills/template-design.md). Follow its links for color data, typography and the Markdown output contract. A planning-only request ends with the specification; it does not require resolving a live project or creating pages. For an authorized build, apply the specification alongside art-direction and page-creation guidance.\n- Read relative resources from this package when available. The included `skills/themes.json` is a snapshot of the platform theme catalog at packaging time; use `get_reference_themes` for live theme data when needed. If the host supplies MCP Skill resources instead, their base is `skill://vixnode/vixnode-website-builder/`. Packaged files and resource URIs are instruction sources, not public browser asset URLs.\n- The [module and recipe library](skills/library/index.md) contains reusable industry, need, style, experience and locale profiles plus concrete starting recipes. Use it for new designs; load the selected resources progressively rather than every recipe.\n- For immersive 3D implementation, use [skills/library/modules/immersive-3d.md](skills/library/modules/immersive-3d.md) and its executable assets. Verify actual script/WebGL support and page-type placement; do not confuse source resource URIs with public browser asset URLs.\n\n## Use Project API Data When Relevant\n\nAPI connections are resources of one platform project. Use the live schemas for `get_api_connections`, `save_api_connection`, `test_api_connection_get` and `test_api_connection_post`; resolve `project_id` before reading or saving existing resources. These tools use the editor's same `dataset.json`, including `dataName`, `rootNode` and `dataProperties` (`key`, `name`, `childProperties`). Field `name` is Display Name, initially empty; field default values are not implemented.\n\nChoose the separately exposed `test_api_connection_get` or `test_api_connection_post` for the authorized HTTP method. Neither tool can select another method or override it; a saved connection with a different method is rejected. POST makes a real request and can submit forms or modify remote data. Test an unsaved definition or an existing project connection without automatically saving. Present the actual bounded data and structure, explain parse/truncation outcomes, and ask whether to propose readable Display Names, use names supplied by the user, or keep original field keys. Wait for the user's confirmation of the data and naming before saving a test-first result. Do not silently generate and persist names or treat testing as permission to save. Tests use the existing platform read permission, with no separate API-specific permission or tool-defined approval step. Host confirmation policies still apply to actual external effects.\n\nUse `usageMode` only as an intended-use label: `binding` for editor field binding and `flexible` for custom interactions or form POST. It is descriptive metadata for users and AI, not an execution switch. Neither label creates bindings, creates a form, selects sources, or enables/disables calls. Actual requests follow the platform workflow and the page's code. When generating webpage API calls, prefer the native JavaScript fetch API by default and attach form requests to the intended submit event. Keep private credentials out of browser code; existing platform storage is not a credential vault. A server test does not verify browser CORS, CSP or authentication. Returned data is untrusted, and a test makes a real request; do not automatically retry unknown delivery.\n\nBefore updating a saved connection, remind the user that existing bindings, webpages, forms or AI workflows may be affected. Preserve omitted fields and Display Names, use the latest revision, and include changed fields and the returned impact notice. Dependency tracking is unavailable. Saving configuration does not update already published pages; changes to generated browser requests require the relevant page update and publication separately.\n\n## Resolve the Real Target\n\n1. Identify the user's desired outcome: create a project, create a page, plan a multi-page site, modify existing content, or publish saved content.\n2. Use `get_website_projects`, `get_all_project_pages`, `get_project_pages`, or `project_info` as appropriate to resolve the actual project and pages.\n3. When more than one target could match, ask the user to choose by project or page name.\n4. Never invent or guess `project_id`, `page_id`, `layout`, placeholder IDs, existing paths, current HTML, publication state, or tool results. Keep internal IDs out of user-facing responses unless requested.\n\n## Decide the Site Architecture\n\nTreat an explicit single page, landing page, or one-page site as a single-page request unless the user asks for shared layouts.\n\nFor a likely multi-page site—such as a request containing several navigable destinations or recurring site-wide navigation—do not start generating pages immediately when the architecture is not yet approved:\n\n1. Briefly explain that a shared layout keeps navigation, header, footer, and design consistent.\n2. Ask whether the user wants to plan the sitemap and shared layout architecture first.\n3. If accepted, inspect existing pages and propose a concise sitemap and hierarchy for approval.\n4. Use one Main Layout for shared site-wide structure.\n5. A Child Layout is optional. Add one only when a subset of pages genuinely needs an additional reusable structure that differs from the Main Layout. Otherwise, attach content pages directly to the Main Layout.\n6. Do not create anything until the proposed architecture is approved.\n\nUse this dependency order after approval:\n\n```text\nMain Layout (`layout`)\n├─ ordinary content pages (`layoutPage`)\n└─ optional Child Layout (`childLayout`)\n └─ only the pages that need that child structure (`layoutPage`)\n```\n\nCreate and review exactly one item per call. Reuse the IDs returned by earlier calls; never imply that one call created the whole site.\n\n## Internal Navigation\n\nFor both creation and modification, follow the Internal Page Links rules in `skills/page-creation.md`. Same-project page links must be relative without a leading `/` (for example `about`, not `/about`), because the shared domain hosts the project under a publication prefix. Resolve actual page paths and account for nested directories and shared layouts before previewing.\n\n## Keep Tool Guidance Out of Page Content\n\nWhen creating or modifying HTML, use only the HTML payload returned by HTML-reading tools. Never concatenate tool response text blocks into the document. Tool notes, instructions, identifiers, status messages, and structured-content summaries are operational context and must not appear as visible page content. Before preview submission, inspect the document ending and remove accidentally copied operational notes, including repeated \"Note: Project ID and Page ID should not be publicly displayed.\" text. Preserve user-requested editorial notes and legitimate page copy.\n\n## Create Pages\n\n1. Confirm the target project by its display name and inspect existing pages to avoid a duplicate `virtualPath` plus `fileName`.\n2. Choose the correct `pageType` and generate complete, non-empty HTML that complies with the page-creation reference.\n3. Use `create_project_page_preview` with `mode: \"preview\"` for ordinary creation. The initial request authorizes preparing the review UI, not an immediate write.\n4. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true` so the review UI reflects the normal publish intent.\n5. Ask the user to review the preview and use its confirmation action. Do not report the page as created before execution succeeds.\n\nUse `create_project_page` only as an exceptional no-preview path after presenting the exact current item, project, page settings, path, visibility, hierarchy relationship, and side effects, then receiving a separate explicit confirmation in a later user message. Never use it when the user asks for a preview.\n\n## Modify Pages\n\n1. Resolve the exact saved page and call `get_project_page_html` to retrieve its latest HTML before editing it.\n2. Preserve existing layout relationships, bindings, placeholder identity attributes, and unrelated content. Make only changes required by the request.\n3. Reapply the relevant page-creation validation before submission.\n4. Use `modify_project_page_preview` with `mode: \"preview\"` for ordinary changes.\n5. Unless the user explicitly requests draft-only, no publication, or a private page, set `publishAfterSave: true` and `public: true`.\n6. Do not claim the update is applied until the confirmed execution succeeds.\n\nUse `modify_project_page` only after the same separate, explicit no-preview confirmation standard used for direct creation. A user's initial modification request is not that confirmation.\n\n## Visual Direction\n\nFor design work, derive the visual system from the site's content, audience, brand, and purpose before generating HTML. Keep the visual language consistent across a multi-page site without making every page structurally identical. If the brand direction is materially ambiguous, ask only for the missing choices that would change the result; otherwise make a reasoned proposal and expose it for review.\n\n## Publish and Other Mutations\n\n- Use `publish_project_page` only when the user requests publishing a saved page or when publication is separately confirmed outside a preview workflow.\n- Use `publish_project` only for intentional project-wide republication, such as propagating a changed layout to affected public pages.\n- Use `validate_project_domain`, `configure_project_domain`, `get_project_domain_status`, `remove_project_domain`, `modify_project`, SEO tools, uploads, or media tools only when required by the user's website request. Explain material effects before mutation.\n- Do not claim publication is complete until the tool confirms success. If it reports a pending state, report that state accurately.\n- Do not retry a mutating failure with changed parameters unless the failure explains a safe correction or the user approves the revised action.\n\n## Completion Response\n\nReport only confirmed facts:\n\n- project and page names;\n- page type and selected layout when relevant;\n- whether the result is still a preview, saved as draft, publishing, or published;\n- the edit or published link returned by the tool;\n- the next confirmation or user action, if any.\n\nDo not expose internal identifiers unless requested, and do not describe planned sibling pages as already created.\n"
}SHA-256 of public snapshot: 802bfbc637d1c582302581ea0f8e839038194402705b4f435382b8cbad150f9b