@scenerok/sdk 0.26.0 → 0.28.0

package/skills/scenerok/references/shaders-filters.md

@@ -21,7 +21,7 @@ A `filter` line grades everything on screen during its window. Add `target: clip
 | `saturation` | Color intensity |
 | `brightness` | Exposure lift/drop |
 
-Use filters as optional finishing tools — do not add by default.
+Choose filters like any other tool (`choosing.md`). A grade (`cdl` for warmth or coolness, `vignette` around 0.3) ties mixed pictures together; effect filters (`glitch`, `chromatic`, `blur`) carry a strong look and fight readable text. Each one's look and clashes are in `repertoire.md`.
 
 ## VFX filters
 

package/skills/scenerok/references/text-styles.md

@@ -2,6 +2,11 @@
 
 `import motion from "@scenerok/basic-animations"` — use with `animate:` on `text` blocks.
 
+Fonts: `font:` is one Google Fonts family at weight 400. Names with a weight such as
+`"Inter Black"` or `"Inter Bold"` (in some catalog examples below) are not families and fall
+back to a default face in renders; use a heavy family such as `"Archivo Black"` instead.
+Type scale, palette and motion timing: `look-and-feel.md`.
+
 ## Motion primitives
 
 | Primitive | Suits | Phase |

package/skills/scenerok/references/vidscript-guide.md

@@ -11,8 +11,8 @@ Contents: Agent quick start · Program structure · Statements (input, output, l
 1. **Import every package function** — use `import xai from "@scenerok/xai"`, `import cf from "@scenerok/cloudflare"`, `import eleven from "@elevenlabs/music"`, and `import motion from "@scenerok/basic-animations"` before calling `xai.*`, `cf.*`, `eleven.*`, or `motion.*`.
 2. **Validate early** — `scenerok validate file.vid` (errors show as `Line N:C — message`).
 3. **Music package syntax** — `import eleven from "@elevenlabs/music"`, then call `eleven.music(...)`, `eleven.generateMusic(...)`, or `eleven.composeMusic(...)`.
-4. **Copy working examples** — `vidscript-sample.md`, and the `.vid` files in the `@scenerok/cli` package's `examples/system/` (e.g. `minimal-text-reel.vid`, `ecom-product-launch.vid`, `rokmilk-chocolate-promo.vid`).
-5. **Use URL assets selectively** — if the user gives a website or product URL and browser screenshot tools are available, you can capture above-fold screenshots, product images, app screenshots, and logos for VidScript `input` assets. Do not treat screenshots as mandatory or as the only visual source.
+4. **Choose per beat** — name each beat's job, then pick tools with `choosing.md` and `repertoire.md` (what each tool looks like, is for, and fights). `recipes.md` has complete scripts; `look-and-feel.md` has the type scale and safe areas.
+5. **Look at a URL's pictures before choosing** — when the user gives a website or product URL, collect the page's own product images, logo and focused screenshots (`look-and-feel.md` → Sourcing imagery), then decide per beat whether to use them. Don't use a full-page screenshot as a background.
 6. **Probe asset type and dimensions** — before placing real assets, identify image vs video with `file`; inspect width, height, and duration with `ffprobe`, `sips`, or `identify`; then scale from the measured aspect ratio and set explicit `width`, `height`, `x`, and `y`.
 7. **Use still images directly** — `input` supports videos and images. Place still screenshots, logos, and product photos with `video image_input`; the renderer treats images as static clips for the block duration.
 8. **Use the full media plugin toolkit** — choose still generation, text-to-video, image-to-video, reference-to-video, video extension, TTS, and music deliberately. Do not default to only `xai.imagine`.
@@ -681,7 +681,7 @@ import motion from "@scenerok/basic-animations"
 [0s .. 5s] = filter "glitch", intensity: 0.5, animate: motion.fadeIn(1s)
 ```
 
-Filters are optional finishing tools; use them only when they improve the composition. Parameter names per filter: `editing-actions.md` → Color and finishing.
+Grade the whole video once (`cdl`, `vignette`, or a `lut`); effect filters are optional accents. Parameter names per filter: `editing-actions.md` → Color and finishing.
 
 ### Animation functions (`@scenerok/basic-animations`)
 

package/skills/scenerok/references/vidscript-strict.md

@@ -15,9 +15,9 @@ Without the owning import, calls like `xai.imagine`, `cf.video`, `xai.tts`, `ele
 
 **Never** call package functions without importing the package alias first.
 
-## Rule 2: Use URL assets selectively
+## Rule 2: Look at a URL's pictures before you choose
 
-A product or landing-page URL is optional source material. Capture focused, legible pieces (above-the-fold views, product shots, logos), never full-page screenshots as backgrounds, and don't invent claims. The full URL workflow is in `workflow.md`.
+When the user gives a product or landing-page URL, collect its own pictures (product shots, logos, focused screenshots) before you decide what each beat shows: they are the only pictures that prove anything about the product. Reading only the title and price leaves you choosing without the options. Use focused, legible pieces, never a full-page screenshot as a background, and don't invent claims. How to collect and prepare them: `look-and-feel.md` → Sourcing imagery.
 
 ## Rule 2.5: Probe real asset type and dimensions before placement
 
@@ -46,7 +46,7 @@ x = (outputWidth - scaledWidth) / 2
 y = safeAreaTop
 ```
 
-Add entry motion to real assets, not only text. When a static visual needs a clean exit, split it:
+Real assets take `animate:` like text does. When a visual should leave softly, split it:
 
 ```vidscript
 import motion from "@scenerok/basic-animations"
@@ -211,7 +211,7 @@ let bed = eleven.music("Soft premium ad music", duration: 15, instrumental: true
 [0s .. 15s] = filter "saturation", amount: 1.12
 ```
 
-Filters are optional finishing tools, not required ingredients. Use them only when they improve the visual result. Each filter has its own parameter name (`saturation` takes `amount`, `blur` takes `radius`); a wrong name is ignored without an error. The table is in `editing-actions.md` → Color and finishing.
+Filters are tools like any other: use a grade (`cdl`, `vignette`, or a `lut` from an HTTPS URL) when the pictures need tying together, and an effect (`glitch`, `chromatic`, `blur`) when its look is the point. What each looks like and fights: `repertoire.md`. Each filter has its own parameter name (`saturation` takes `amount`, `blur` takes `radius`); a wrong name is ignored without an error. The table is in `editing-actions.md` → Color and finishing.
 
 A `filter` line grades everything in its window. To grade one clip, add `target: clip` or write `clip.Filter("sepia", intensity: 0.6)`.
 

package/skills/scenerok/references/workflow.md

@@ -2,35 +2,29 @@
 
 ## Pipeline
 
-1. **Understand the goal** — promo, testimonial, meme, code demo, etc.
+1. **Understand the goal** — promo, testimonial, meme, code demo, etc. Pick the closest script in `recipes.md`
 2. **Choose authoring** — VidScript (`.vid`) by default and for every editing action; TypeScript `@scenerok/sdk` (`.ts`, `sdk-guide.md`) only for loops or computed timing
-3. **Plan the structure** — time blocks, durations, inputs
-4. **Gather assets** — URLs, local paths, generated clips, selective website screenshots
-5. **Probe asset type and dimensions** — `loadVideo` / media stdlib, or `file` / `ffprobe` / `sips` / `identify`; never guess
-6. **Use images directly when useful** — VidScript `video image_input`; SDK `video(path)` / `loadVideo`
+3. **Plan the structure** — beats of 2.5–4.5 s, each with a picture, copy, and motion (`look-and-feel.md`)
+4. **Put the material on the table** — page images, focused screenshots, logo, footage, screen recordings, the exact numbers and words. Generated stills or clips when nothing real exists and the beat needs a picture. You can't choose a picture you never looked at
+5. **Probe and prepare** — `file` / `ffprobe` / `sips` / `identify` (or `loadVideo` / media stdlib); crop and blur stills with ffmpeg; never guess sizes
+6. **Decide the look** — palette, two fonts, and type sizes as `let` values (`look-and-feel.md`)
 7. **Choose plugin modality** — still / t2v / i2v / ref2v / extend / TTS / music (`ai-models.md`, `tts.md`, `music.md`)
 8. **Compose** — measured dimensions, `area:` / `t.grid` for text, symbolic durations (`vo.duration` / returned node `.duration`), entry/exit beats
 9. **Validate** — `scenerok validate script.vid` or `scenerok validate scene.ts`
 10. **Deep-validate (VidScript)** — for scripts with TTS/generated media: `scenerok validate script.vid --generate-assets --report-json` (billed; assets cache for the render). SDK plugins generate during `scenerok render` instead. Fix `offCanvas`/`wasShrunk` text, `collisions`, and unintended `audio.overlaps`, then re-validate.
 11. **Render** — `scenerok render script.vid --watch` or `scenerok render scene.ts --watch`
-12. **Check the render** — `scenerok lint-video <render-id> --cost medium --json` (billed); fix and re-render if it reports problems. `scenerok render script.vid --validate` does render and check in one step
-13. **Deliver** — share download URL when complete
+12. **Look at the render** — pull one frame per beat with ffmpeg, open them, and answer the review questions in `SKILL.md`. Fix what you see and render again. `scenerok lint-video <render-id> --cost medium --json` (billed) adds an LLM review of frame grabs
+13. **Deliver** — share the download URL once you can defend each beat
 
 ## Website URL asset workflow
 
-When the user provides a product, company, landing page, or ecommerce URL:
-
-1. Visit the URL with a browser tool before composing.
-2. Capture screenshots sparingly — above-fold, product closeups, pricing cards. Avoid full-page backgrounds.
-3. Crop to the visual evidence needed.
-4. Save product images, logos, app screenshots when accessible.
-5. Probe saved asset dimensions before composing.
-6. Use stills as timeline inputs or feed into `xai.imageToVideo` / `xai.referenceToVideo` (VidScript or SDK).
-7. Combine real assets with generated clips, text, and music.
-8. Scale from measured aspect ratio; keep primary visuals in ~80–88% frame width for 9:16.
-9. Animate assets: `motion.popIn` / SDK `popIn`, `fadeIn`, `slideY`.
-10. Generative prompts must require **no text, logos, watermarks, or readable UI** in AI output — add copy with `text`.
-11. Do not invent product claims — extract from site or ask the user.
+When the user gives a product, company, landing-page or store URL, the page's
+own pictures go in the video. Reading only the title, price and bullets is not
+enough. Collect the image URLs and focused screenshots, probe them, crop out
+baked-in captions, and place them as cards, bands over a blurred fill, or
+full-frame covers: `look-and-feel.md` → Pictures and Sourcing imagery. A worked
+example is the product promo in `recipes.md`. Don't invent product claims;
+extract them from the page or ask.
 
 ## Best practices
 
@@ -62,7 +56,9 @@ If render fails: check `scenerok status <id>`, fix missing assets or syntax, re-
 
 ## Sample video types
 
-- **Product promo** — hero + headline + CTA
+- **Product promo** — hero image, features, price + CTA (`recipes.md`)
+- **Release notes** — logo, typed code, stat, outro (`recipes.md`)
+- **Talking head** — speaker cut, punch-ins, b-roll, captions (`recipes.md`)
 - **Social hook** — fast cuts, bold text
 - **Testimonial** — speaker clip + quote overlay
 - **Meme remix** — reaction clip + punchlines