minutework 0.1.56 → 0.1.58

package/assets/claude-local/CLAUDE.md.template

@@ -73,6 +73,26 @@ When the task involves event-driven behavior or reacting to external/reference
 datasets, read `skills/event-bus/SKILL.md` and
 `skills/cross-server-subscriptions/SKILL.md`.
 
+## Workspace Customizations (Authored Overlays)
+
+This workspace may carry developer-authored customizations that sit ALONGSIDE
+the managed guidance and are never managed, overwritten, or refreshed by
+`minutework workspace sync-assets`:
+
+- `CLAUDE.workspace.md` / `AGENTS.workspace.md`: workspace-specific guidance that
+  extends this file. In Claude Code, read `CLAUDE.workspace.md` alongside
+  `CLAUDE.md`; Codex and Cursor read `AGENTS.workspace.md` alongside `AGENTS.md`.
+  If the file is present, follow it the same way you follow this guidance.
+- `skills.workspace/`: workspace-authored skills, each a `SKILL.md` with the same
+  shape as the managed `skills/`. Browse `skills.workspace/` alongside `skills/`
+  and read the relevant `SKILL.md` when a topic matches your task.
+
+These authored files are intentionally outside the managed asset manifest, so
+`sync-assets` can update the managed baseline (`CLAUDE.md`, `AGENTS.md`,
+`skills/`) without ever touching your customizations. Authored overlays EXTEND
+this managed guidance for the specific workspace; they do not relax the Hard
+Rules, hard boundaries, or trust boundaries stated above.
+
 ## Capability Gaps
 
 When a request exposes missing shared MinuteWork substrate, read

package/assets/claude-local/skills/README.md

@@ -54,6 +54,8 @@ Generated-workspace-first guidance should live here, especially:
 - `vuilder-workspace-architecture/SKILL.md`
 - `vuilder-public-site-authoring/SKILL.md`
 - `vuilder-mode-b-hosted-and-spawn/SKILL.md`
+- `visualize-context/SKILL.md`
+- `landing-page-assets/SKILL.md`
 - `workspace-guidance-refresh/SKILL.md`
 - `shell-architecture/SKILL.md`
 - `runtime-capability-inventory/SKILL.md`

package/assets/claude-local/skills/landing-page-assets/SKILL.md

@@ -0,0 +1,153 @@
+---
+name: landing-page-assets
+description: "Generate website-ready hero images, backgrounds, section visuals, textures, and OG assets, then wire selected images into public-site surfaces."
+---
+
+# Landing Page Assets
+
+Use this skill when the user asks for website-bound imagery, for example:
+
+- "generate hero image"
+- "background image"
+- "section below it"
+- "website asset"
+- "landing page visual"
+- "use this in the site"
+- "OG image"
+- "texture"
+
+Default v1 output: generate an actual image, store it, add alt text, and wire
+the selected asset into the target site/page/manifest when a target surface is
+available.
+
+## Asset Roles
+
+Choose one role before prompting:
+
+- `hero_background`: wide atmospheric or metaphor image behind/around HTML text.
+- `hero_visual`: primary product/domain visual next to or within the hero.
+- `section_illustration`: visual supporting a feature/content section.
+- `background_texture`: text-free repeatable or full-bleed texture/motif.
+- `cta_visual`: closing section image that reinforces action.
+- `og_image`: social preview image; text is allowed only when explicitly needed.
+
+Strong rule: prefer HTML text over text baked into images. Only generate a
+poster/graphic with text when the user explicitly asks for a graphic image or
+OG card.
+
+## Workflow
+
+1. Infer the asset role and target section.
+   - Use the latest user request, current page JSON, section id, and prior
+     visual direction.
+   - If a target page/manifest is available, note where the asset should land.
+   - If the user only wants exploration, generate a preview artifact and do not
+     mutate the site.
+
+2. Preserve brand and layout inputs.
+   - Keep palette, typography mood, material style, photo/illustration style,
+     and metaphor constraints.
+   - For verticals like trucking, healthcare, legal, or finance, use grounded
+     domain visuals and avoid generic SaaS abstractions.
+
+3. Write an asset prompt.
+   - Include role, section purpose, subject, composition, palette, lighting,
+     texture, crop safety, responsive layout intent, and negative constraints.
+   - For backgrounds, explicitly request no text, no UI, no logos, and safe
+     empty space where page HTML will sit.
+   - For section images, avoid fake product screenshots unless the user asks
+     for an interface mockup.
+
+4. Generate through `media.generate_image`.
+   - `use_case`: `landing_page_asset`.
+   - Use 16:9 or 21:9 for hero/backgrounds, 4:3 for section visuals, and
+     1200x630 for OG images.
+   - Attach generated outputs to the thread for preview.
+
+5. Promote the selected asset when it belongs in a public site.
+   - Use the public-dj media asset upsert endpoint/service to create or update
+     `MediaAsset` by `vertical + asset_key`.
+   - Store durable metadata: `asset_key`, `digest`, `url`, `alt_text`, `role`,
+     dimensions, mime type, and prompt summary.
+   - Do not store long-lived presigned URLs or storage secrets in manifest
+     metadata.
+
+6. Wire the page/manifest.
+   - Add or update manifest `payload.media[]` entries:
+
+```json
+{
+  "asset_key": "fleet-hero-owner-operator",
+  "digest": "sha256:...",
+  "url": "/api/v1/media/fleetrun/fleet-hero-owner-operator/sha256:...",
+  "alt_text": "Owner-operator reviewing an insurance quote beside a semi truck",
+  "role": "hero_background",
+  "metadata": {
+    "width": 1600,
+    "height": 900,
+    "mime_type": "image/png"
+  }
+}
+```
+
+   - In section payloads, reference images by `asset_key` when possible:
+
+```json
+{
+  "type": "hero_with_cta",
+  "headline": "Instant quote. Instant answers.",
+  "image": {
+    "asset_key": "fleet-hero-owner-operator",
+    "role": "hero_background"
+  }
+}
+```
+
+   - Supported section image fields for v1:
+     `hero_with_cta.image`, `feature_grid.items[].image`,
+     `content_markdown.image`, and `cta_banner.image`.
+   - For workspace-only generated assets, place selected files under the
+     site's public asset directory and reference them directly.
+
+## Prompt Patterns
+
+Hero/background:
+
+```text
+Create a website hero_background for [product/page] serving [audience].
+Subject: [domain-specific scene or metaphor].
+Composition: wide crop, responsive-safe, clear negative space for HTML text on [side].
+Style: [brand style, palette, lighting, realism level].
+No text, no UI, no logos, no fake brand marks.
+Aspect ratio: 16:9 or 21:9.
+```
+
+Section visual:
+
+```text
+Create a section_illustration for the section "[section headline]".
+Show [specific concept/user/workflow] in a grounded way.
+Composition: uncluttered, works beside HTML text, no tiny labels.
+Style: [brand cues].
+Avoid generic SaaS dashboards unless a product UI is explicitly requested.
+```
+
+OG image:
+
+```text
+Create an og_image for [product/page].
+Use the brand/product as the first-viewport signal.
+Readable text only if explicitly requested; otherwise leave text to HTML/social metadata.
+Aspect ratio: 1200x630.
+```
+
+## Guardrails
+
+- Do not bake landing page body copy into imagery.
+- Do not generate fake customer logos, certifications, insurance carrier logos,
+  DOT records, prices, or legal/compliance claims.
+- Do not use dark, blurred, cropped, stock-like imagery when the asset needs to
+  communicate the actual product, buyer, or domain.
+- Do not promote an asset into a site unless the user asked to use it or the
+  current task is site generation.
+- Always provide alt text for promoted assets.

package/assets/claude-local/skills/runtime-capability-inventory/SKILL.md

@@ -135,6 +135,9 @@ generated Builder workspace.
     action capability; preflight fail-closes on connection/secret availability, and the
     delivery row is idempotent on the ActionRun id with a SENT short-circuit. Not
     agent-exposed yet: exposure is a separate per-definition mw.core.agent_actions flip.
+  - `media.generate_image` -- Runtime media action capability wraps the existing
+    generate_image service, stores outputs in runtime_storage, and attaches
+    generated_image thread artifacts when thread context is present.
   - `consent.check_channel` -- Service seam consumed by other executors and sidecars as
     a preflight; not an agent-callable tool.
   - `discovery.request_proposal` -- Effectful; outside the default READ_RING, granted

package/assets/claude-local/skills/runtime-capability-inventory/primitive-catalog.json

@@ -217,6 +217,25 @@
       "approval_required": false,
       "interim_path": ""
     },
+    {
+      "primitive_id": "media.generate_image",
+      "family": "media",
+      "display_name": "Generate Image",
+      "description": "Generate image assets through the runtime media provider and store outputs as runtime files.",
+      "runtime_app": "runtime_media",
+      "dispatch_path": "",
+      "status": "active",
+      "requires_integration": false,
+      "requires_secret": true,
+      "metered": true,
+      "executor_ready": "live",
+      "executor_ref": "capability:media.generate_image",
+      "executor_notes": "Runtime media action capability wraps the existing generate_image service, stores outputs in runtime_storage, and attaches generated_image thread artifacts when thread context is present.",
+      "agent_exposable": true,
+      "agent_exposure_path": "app_pack_action",
+      "approval_required": false,
+      "interim_path": ""
+    },
     {
       "primitive_id": "consent.check_channel",
       "family": "consent",

package/assets/claude-local/skills/visualize-context/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: visualize-context
+description: "Generate strategy, product, system, PLG, and concept visuals from messy context, then attach generated images to the Builder thread."
+---
+
+# Visualize Context
+
+Use this skill when the user asks to visualize strategy/product context, for
+example:
+
+- "visualize this"
+- "generate an image"
+- "show the flow"
+- "make this easier to see"
+- "PLG loop"
+- "system map"
+- "architecture visual"
+- "background metaphor"
+
+Default v1 output: an actual generated image attached back to the thread as a
+runtime-stored `generated_image` artifact. Do not stop at a prompt unless image
+generation is unavailable.
+
+## Workflow
+
+1. Extract the decision context.
+   - Identify the product, audience, buyer, loop, system, or strategic argument.
+   - Preserve named entities, constraints, data flow, and any stated design style.
+   - Separate verified facts from interpretation. Do not invent traction,
+     customers, logos, certifications, or claims.
+
+2. Choose the visual type.
+   - `product_flow_map`: activation, value moment, retention, conversion.
+   - `plg_loop`: acquisition, free value, habit, share/expand, conversion.
+   - `system_diagram`: components, data movement, approvals, runtime boundaries.
+   - `strategy_map`: thesis, unfair advantage, risks, market timing, proof.
+   - `concept_visual`: metaphor image that makes an abstract idea graspable.
+   - `landing_hero_mockup`: first-screen marketing visual preview only. If the
+     user wants assets used in a site, switch to `landing-page-assets`.
+
+3. Write a structured image prompt.
+   - Put the core visual in the first sentence.
+   - Include layout, hierarchy, style, palette, target aspect ratio, and what
+     should be legible.
+   - Use labels sparingly. Prefer large, readable labels over dense tiny text.
+   - If the image is meant to explain a complex flow, ask for a clean diagram
+     composition, not a photorealistic scene.
+
+4. Generate through `media.generate_image`.
+   - Use `use_case`: `visualize_context`.
+   - Use 16:9 or 4:3 for strategy diagrams; 1:1 only when the user asks for a
+     square artifact.
+   - Set `request_source_ref` to the current thread/run context when available.
+   - Attach the generated output as a thread artifact. The action capability
+     does this automatically when thread context is present.
+
+5. Respond with the result.
+   - Explain the visual in one or two sentences only if needed.
+   - Do not claim the image proves anything. It is a visualization aid.
+
+## Prompt Contract
+
+For visual prompts, include:
+
+- `visual_type`
+- `main_subject`
+- `audience`
+- `flow_or_components`
+- `style_reference`
+- `brand_constraints`
+- `must_show`
+- `must_avoid`
+- `aspect_ratio`
+
+Use this shape internally:
+
+```text
+Create a [visual_type] showing [main_subject] for [audience].
+Composition: [layout and hierarchy].
+Content: [flow/components, 4-7 max].
+Style: [brand/design cues].
+Readable text: [short labels only].
+Avoid: [fake logos, claims, tiny text, generic stock visuals].
+Aspect ratio: [16:9 / 4:3 / square].
+```
+
+## Guardrails
+
+- Do not invent fake customer names, fake logos, fake metrics, fake compliance,
+  or market claims.
+- Do not create tiny unreadable UI text. Use simple labels and preserve detail
+  in the thread answer or source content.
+- Do not bake long paragraphs into images.
+- Do not use generic SaaS stock visuals when the user supplied a specific
+  domain, metaphor, or brand style.
+- If the user asks for a website asset or says "use it in the site", switch to
+  `landing-page-assets` because that needs asset role, alt text, and promotion.