@hraness/slopcamera 3.3.6 → 3.5.0

package/skills/slopcamera/references/brand-illustrations.md

@@ -0,0 +1,237 @@
+# Brand illustrations
+
+Read this before you generate, trace, retouch, recolor, resize or ship any
+product illustration or topic icon for a marketing page, docs landing, project
+card or social card. A trace that "looks fine" at 448 px is not evidence. The
+gate below is.
+
+Scope: the **illustration** class of Slopcamera's two-class product artwork
+(`slopcamera image icon --purpose illustration`), meaning marketing and topic
+art shown at 24 to 160 px on Hraness sites. Marks (`--purpose mark`) keep their
+own contract. Interface icons (nav, buttons, labels) stay on the stroke sprite
+algal.computer uses (hugeicons, 24 viewBox, stroke 1.5, currentColor).
+
+### The icon language
+
+Every illustration in the portfolio is one family. Product identity comes from
+the subject and the palette primary the site applies. It does not come from
+per-product line weight, per-product ink or a different drawing style.
+
+| Property | Rule |
+| --- | --- |
+| Frame | `viewBox="0 0 64 64"`, square. The subject's long side is 48 to 54 units, optically centered. The 5-unit margin is empty. |
+| Line | Monoline centerline strokes: `fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"`. |
+| Weights | Four tones and nothing else: **line** 2 u (the body of the drawing); **detail** 1.5 u (interior detail only, at most 25% of total line length); **accent** 3 u (at most one element, the focal point); **tint plane** `fill="currentColor" fill-opacity=".16"` (at most 2 paths, on the primary face only). No hairlines under 1.5 u. No solid fills except a tint plane. |
+| Projection | Simple 30-degree isometric, or front orthographic for flat subjects such as a page, card or screen. Use one projection per site set. |
+| Subject | One main object and at most one small secondary prop. At most 8 connected parts. No text, digits, letters, logos, hatching, texture, shading, gradients, shadows, ground planes, badges, frames or container tiles. |
+| Spacing | Parallel strokes are at least 3 u apart (1.5 times the line). Closed counters are at least 6 u across. No stroke ends closer than 2 u to another stroke unless they join. |
+| Color | `currentColor` only. No hex, rgb, named color, `<style>`, media query or `class` paint in the file. The site sets `color` (normally `var(--primary)`). |
+| Markup | Only `<svg>`, `<g>`, `<path>`, `<circle>`, `<ellipse>`, `<rect>`, `<line>`, `<polyline>`. At most 24 elements and 400 path commands, 0.1-unit precision, 6 KB or less. No `mask`, `clipPath`, `filter`, `transform`, `vector-effect`, `use`, `image`, `style`, `script`, `href` or `url()`. No `data-slopcamera-line-weight`. |
+
+#### Optical sizes
+
+Stroke weight is fixed in frame units, so rendered weight scales with display
+size. Three optical sizes keep the rendered line between about 1.2 and 3 CSS px,
+roughly the square root of the size ratio to algal's 1.2 to 1.3 rem accent icons
+(1.5/24 stroke gives 1.2 px at 19.2 px):
+
+| Optical size | Display | Stroke | Detail length | Parts | Tint plane | File |
+| --- | --- | --- | --- | --- | --- | --- |
+| `s` | 24 to 40 px | 3 u (1.1 to 1.9 px) | 100 to 280 u | 4 or fewer | none | `<id>.s.svg` |
+| `m` (default) | 44 to 120 px | 2 u (1.4 to 3.75 px) | 160 to 460 u | 8 or fewer | 1 or fewer | `<id>.svg` |
+| `l` | 128 px and up | 1.5 u (3 px and up) | 200 to 640 u | 10 or fewer | 2 or fewer | `<id>.l.svg` |
+
+Below 24 px, use the product **mark**, never an illustration. Derive `s` from
+`m` by pruning, never by scaling: drop detail strokes, merge near-parallel
+lines, then re-stroke at 3 u.
+
+#### Delivery
+
+Sites render illustrations so the palette reaches them. Never ship them through
+a bare `<img>` with a baked color.
+
+1. **Preferred:** a generated per-site sprite, `/illustrations.svg`, with one
+   `<symbol id="<id>" viewBox="0 0 64 64">` per illustration, rendered as
+   `<svg class="hraness-illustration" data-size="m" aria-hidden="true"><use href="/illustrations.svg#<id>"/></svg>`.
+   `currentColor` inherits through `<use>`, the tint plane keeps its opacity,
+   and forced-colors mode works natively. This is the same sprite pattern
+   algal.computer uses for interface icons (`/icons.svg`).
+2. **Fallback** where only an image URL is possible, such as Markdown or a CMS:
+   the design-kit mask utility (`.hraness-illustration` with a painted
+   mask layer) paints `var(--hraness-illustration-ink, var(--primary))`
+   through the SVG's alpha.
+   It follows the same structure as `.hraness-foil-mark`.
+3. Size and color come from design-kit classes (`.hraness-illustration[data-size="s|m|l"]`).
+   Sites must not own a `.x-topic-icon { width: 88px … }` rule, an `opacity`,
+   a `filter: hue-rotate()` or an `invert()`.
+4. Contrast: `--primary` against the surface the illustration sits on must
+   reach at least 3:1 (WCAG 1.4.11, non-text) in both schemes. Where a card
+   surface is tinted, set `--hraness-illustration-ink` to the palette's
+   strong-primary token instead of lowering opacity.
+
+### Generate
+
+```sh
+slopcamera image icon '<subject>' --purpose illustration \
+  --concept '<one-sentence metaphor>' --size m --rounds 2 --output icons/<id>.svg --json
+slopcamera image icon normalize icons/<id>.svg --size s --output icons/<id>.s.svg --json
+slopcamera image icon check icons/ --json      # gate every file; non-zero exit on any failure
+slopcamera image icon sheet icons/ --palette <family> --output review.png   # 32/64/88 px, light+dark
+```
+
+(`--concept`, `--size`, `normalize`, `check` and `sheet` are the proposed
+additions. Today only `image icon` exists.)
+
+1. **Concept first.** Turn the product verb into one physical object metaphor,
+   in one sentence. Write three candidates. Generate a gallery
+   (`slopcamera image gallery`) of the best two before spending icon rounds.
+   Reject metaphors another portfolio product already owns (see the shared
+   pool in the registry) unless you deliberately reuse the pooled file.
+2. **Raster.** The prompt template below fixes ink, weight ratio, projection
+   and element budget. The model draws black on white. Color is never a
+   generation parameter.
+3. **Normalize** (deterministic, local, no network): centerline re-trace to
+   2 u strokes, strip color to `currentColor`, reframe to 64, emit, then
+   re-measure. See "Normalization" below.
+4. **Gate.** Hard numeric gate (below). A failure feeds a concrete correction
+   into the next round ("remove the 6 shelf slats; keep 2"), not a restyle.
+5. **Critique.** The vision reviewer sees the **sheet**: 32, 64 and 88 px on
+   the target palette's light and dark surfaces, beside the site's other
+   illustrations, with the gate metrics printed underneath. A pass at 448 px
+   alone does not count.
+6. **Register.** Record the file in the artwork registry with subject,
+   concept, sha256, gate version and metrics. Sites consume registry output.
+   They do not keep hand-copied SVGs.
+
+#### Prompt template
+
+```text
+A single product-brand line illustration of {subject}, shown as {concept}.
+
+Drawing rules:
+- Pure black (#000000) lines on a plain white (#FFFFFF) background. No other color, no gray.
+- Every line has the same thickness, about 3% of the image width (about 32 px on a 1024 px image).
+  No thin detail lines. No filled or solid areas{accent_clause}.
+- Simple 30-degree isometric view of one main object{prop_clause}. At most {max_parts} separate parts.
+  Draw as few lines as you can. Every line must still be clearly visible when the image is 64 px wide.
+- Closed shapes are outlines, not filled. Leave at least one line-width of white between parallel lines.
+- Round line ends and rounded corners.
+- No text, numbers, letters, logos, hatching, texture, shading, gradients, shadows, floor, frame,
+  badge, tile or background shape.
+- The object fills about 80% of the image and is centered.
+{feedback_clause}
+```
+
+- `accent_clause`: empty by default. For a deliberate focal accent, use
+  `, except one small solid {part} no larger than a tenth of the drawing`.
+  The normalizer converts it to the tint plane.
+- `prop_clause`: empty, or `, with one small {prop} beside it`.
+- `max_parts`: 8 for `m`, 4 for `s`, 10 for `l`.
+- `feedback_clause`: `The previous attempt was rejected. Change only this: {gate reason codes rendered as concrete edits}`.
+
+Do not ask for brand color, "medium weight", "clean", "minimal" or "premium".
+Those words do not constrain the raster. The numbers do.
+
+#### Normalization
+
+`slopcamera image icon normalize` makes weight and color properties of the
+normalizer, not the image model:
+
+1. Parse the SVG and reject forbidden content (list in "The icon language").
+2. Rasterize its alpha on a **transparent** background, with the art's bounding
+   box fitted into 54 u inside a 64 u frame at 8 px/unit (512 by 512, pinned
+   librsvg/sharp).
+3. Binarize at alpha 128 or more. Compute the Euclidean distance transform and a
+   Zhang–Suen skeleton.
+4. Separate **masses** (ink more than 2 u from any background pixel, meaning a
+   region wider than 4 u). Keep at most 2 masses: trace their outer contours
+   with marching squares and emit them as tint planes
+   (`fill="currentColor" fill-opacity=".16"`). If masses exceed 12% of ink,
+   stop with `mass.excess`: the art is mark-weight and must be regenerated.
+5. Turn the skeleton into a graph (junction and end pixels are nodes). Prune
+   spurs shorter than 2 u. Merge parallel chains closer than 3 u. Remove
+   components smaller than 1.5 u².
+6. Trace each edge chain into a polyline, simplify it (Ramer–Douglas–Peucker
+   with epsilon 0.3 u), fit centripetal Catmull–Rom curves to cubic Béziers
+   with corner detection above 35 degrees, and join chains that meet at a node
+   of degree 2.
+7. Emit at most 24 paths on `viewBox="0 0 64 64"` with the root attributes
+   `fill="none" stroke="currentColor" stroke-width="{2|3|1.5}" stroke-linecap="round" stroke-linejoin="round"`,
+   rounded to 0.1 u. Chains whose source width was under 1.8 u become detail
+   strokes (1.5 u, `m` and `l` only). One chain marked by the concept as the
+   accent becomes 3 u.
+8. Re-render the emitted SVG and run the gate. Normalization never loosens a
+   threshold to pass.
+9. Write a receipt: source sha256, output sha256, gate version, every metric
+   and the pruning counts.
+
+Sketch (TypeScript, Bun; reuses Slopcamera's pinned sharp and has no new
+native dependency):
+
+```ts
+export async function normalizeIllustration(svg: string, size: "s" | "m" | "l"): Promise<NormalizedIllustration> {
+  assertAllowedMarkup(svg)                                   // forbidden tags/attrs/url()
+  const alpha = await renderAlphaFramed(svg, { frame: 64, art: 54, pxPerUnit: 8, background: "transparent" })
+  const ink = binarize(alpha, 128)
+  const dist = euclideanDistance(ink)                        // Felzenszwalb–Huttenlocher
+  const masses = regions(ink, (i) => dist[i] > 2 * 8)
+  if (share(masses, ink) > 0.12) throw gateError("mass.excess")
+  const graph = skeletonGraph(zhangSuen(ink))
+  prune(graph, { spurUnits: 2, mergeParallelUnits: 3, minComponentUnits2: 1.5 })
+  const chains = graph.chains().map((c) => fitCubic(rdp(c.points, 0.3 * 8), { cornerDeg: 35 }))
+  const out = emitSvg({
+    viewBox: "0 0 64 64",
+    stroke: { s: 3, m: 2, l: 1.5 }[size],
+    chains: classify(chains, dist),                          // line | detail | accent
+    tints: masses.slice(0, size === "s" ? 0 : size === "m" ? 1 : 2).map(marchingSquares),
+    precision: 0.1,
+  })
+  const metrics = await measureIllustration(out)             // same code path as `icon check`
+  const problems = illustrationGate(metrics, size)
+  if (problems.length) throw gateError(problems)
+  return { svg: out, metrics, receipt: receiptFor(svg, out, metrics) }
+}
+```
+
+Re-stroking rescues drawings whose composition is sound but whose weight was
+set by the image model. It cannot rescue mass art or over-detailed art; those
+need regeneration from the prompt template with a smaller element budget.
+
+#### Gate (version 1)
+
+All metrics come from the framed render in normalization step 2, identical in
+`image icon`, `icon check`, the jungle registry and CI.
+
+| Code | Metric | Pass band (`m`) | `s` | `l` |
+| --- | --- | --- | --- | --- |
+| `weight.faint` / `weight.heavy` | median stroke width at skeleton points, (2·EDT − 1)/8 | 1.7 to 2.4 u | 2.6 to 3.4 u | 1.3 to 1.9 u |
+| `weight.hairline` | share of skeleton under 1.0 u wide | 5% or less | 3% or less | 8% or less |
+| `weight.mixed` | p90 stroke width while mass is above 2% | 3.0 u or less | 3.8 u or less | 3.0 u or less |
+| `mass.excess` | ink more than 2 u from background, as share of ink | 12% or less | 0% | 12% or less |
+| `coverage.sparse` / `coverage.dense` | Σα / frame area (transparent letterbox) | 0.09 to 0.22 | 0.08 to 0.20 | 0.07 to 0.22 |
+| `detail.thin` / `detail.crowded` | skeleton length | 160 to 460 u | 100 to 280 u | 200 to 640 u |
+| `parts.fragmented` | 8-connected components of 1 u² or more | 8 or fewer | 4 or fewer | 10 or fewer |
+| `parts.specks` | components under 1.5 u² | 0 | 0 | 0 |
+| `frame.aspect` | art bounding-box aspect | 1.6 or less | 1.4 or less | 1.8 or less |
+| `svg.frame` | viewBox | `0 0 64 64` | same | same |
+| `svg.color` | literal paint present, or `currentColor` absent | none / present | same | same |
+| `svg.markup` | forbidden element or attribute | none | same | same |
+| `svg.paths` / `svg.bytes` | element count / file size | 24 or fewer / 6 KB or less | 12 or fewer / 3 KB or less | 32 or fewer / 9 KB or less |
+
+The bands hang together. At a 2 u line, 160 to 460 u of drawn length is 320 to
+920 u² of ink, which is 0.08 to 0.22 of the 4096 u² frame. Coverage and detail
+length therefore agree unless the art contains masses or hairlines, and those
+have their own codes.
+
+#### Review checklist
+
+Run on the sheet, not on a single large render:
+
+- [ ] Recognizable at 32 px without its label, on both light and dark surfaces.
+- [ ] Line weight matches its neighbors on the site sheet. No drawing reads fainter or bolder than the rest.
+- [ ] No stroke, gap or counter is narrower than the line.
+- [ ] At most one accent and one tint plane, both on the focal element.
+- [ ] Renders in the site's `--primary` through the sprite or mask, not in `#2474d4`. Contrast is 3:1 or better on the actual card surface.
+- [ ] The concept matches the copy of the section it heads, and differs from other products' concepts unless it is pooled.
+- [ ] `s` exists wherever the site shows the illustration at 40 px or less.
+- [ ] `icon check` passes and the registry receipt is updated. There is no hand-edited copy in any site.

package/skills/slopcamera/references/gateway-media.md

@@ -6,6 +6,11 @@ credential. It has no Slopcamera account, credential store, or hosted project.
 
 ## Start with the intended role
 
+For a creative film or animation brief, use
+[visual style direction](visual-style-direction.md) to translate the intended
+look into concrete source, camera, movement, and finishing decisions. Keep a
+literal prompt literal when that is what the user requested.
+
 Clarify what the generated media needs to do in the final project:
 
 - the subject or message;
@@ -131,7 +136,7 @@ Use [directing video](directing-video.md) for budgeted takes, accepted predecess
 
 The small portable `slopcamera image generate '<prompt>' --output image.webp` command defaults to `recraft/recraft-v4.1-utility` and admits PNG/JPEG/WebP output. It is distinct from the local `ai` catalog workflow. Preserve literal prompts when requested, keep its output inside the intended workspace and never automatically retry a failed paid call. Its sibling `slopcamera image gallery '<subject>' --output-dir <directory>` shares that bounded model contract and composes several candidates into one review sheet; see [image galleries](image-galleries.md).
 
-For product identity SVGs, `slopcamera image icon '<subject>' --purpose <mark|illustration> --output icon.svg` is the canned pipeline. `mark` produces a compact favicon/app/header mark: one to three bold masses in one ink, gated for 16–32 px recognition and low path count. `illustration` (the compatibility default) produces the related marketing illustration: simple isometric structure in one ink, gated against hairlines, dense fills, and detail lost at 64 px. Both run one style-locked Gateway raster through ink normalization, local tracing, and deterministic geometry gates. `--rounds 2` (the default) adds a purpose-specific vision critique that revises the prompt between attempts; `--rounds 1` is a single paid generation with no critique call. `--ink` overrides the ink color and `--keep-raster` retains the normalized line-art PNG beside the SVG.
+For product identity SVGs, `slopcamera image icon '<subject>' --purpose <mark|illustration> --output icon.svg` is the canned pipeline. `mark` produces a compact favicon/app/header mark: one to three bold masses in one ink, gated for 16–32 px recognition and low path count. `illustration` (the compatibility default) produces the related marketing illustration: simple isometric structure in one ink, gated against hairlines, dense fills, and detail lost at 64 px. Both run one style-locked Gateway raster through ink normalization, local tracing, and deterministic geometry gates. `--rounds 2` (the default) adds a purpose-specific vision critique that revises the prompt between attempts; `--rounds 1` is a single paid generation with no critique call. `--ink` overrides the ink color and `--keep-raster` retains the normalized line-art PNG beside the SVG. For portfolio illustrations and topic icons, follow [brand illustrations](brand-illustrations.md): one shared line language, `currentColor` output, measured weight and density bands, and the site sheet review.
 
 ## Style recipes
 

package/skills/slopcamera/references/install.md

@@ -14,7 +14,16 @@ command -v slopcamera
 
 A restricted shell can omit package-manager paths. Check known host installation paths before declaring a tool unavailable. If Bun is genuinely absent, follow its official [installation guide](https://bun.sh/docs/installation) within the user’s authorized setup scope. Do not switch package managers or pipe an unreviewed installer into a shell.
 
-Slopcamera installs from its verified release archive: `bun add --global https://github.com/hraness/slopcamera/releases/download/v3.3.5/hraness-slopcamera-3.3.5.tgz`, then `slopcamera doctor --json`. Building from source is the contributor path. Historical Atet archives do not install the renamed CLI; never substitute `slopcamera` into an old archive URL.
+Slopcamera installs from its verified release archive: `bun add --global https://github.com/hraness/slopcamera/releases/download/v3.4.0/hraness-slopcamera-3.4.0.tgz`, then `slopcamera doctor --json`. Building from source is the contributor path. Historical Atet archives do not install the renamed CLI; never substitute `slopcamera` into an old archive URL.
+
+If an update from a previous local archive reports `DependencyLoop`, use Bun's
+[named tarball syntax](https://bun.com/docs/pm/cli/add#tarball-dependencies) with
+the same verified archive:
+
+```sh
+bun add --global @hraness/slopcamera@https://github.com/hraness/slopcamera/releases/download/v3.4.0/hraness-slopcamera-3.4.0.tgz
+slopcamera --version
+```
 
 Use an existing compatible source build when available. Otherwise follow the [complete source-install guide](https://github.com/hraness/slopcamera/blob/main/docs/how-to/use-current-source.md): clone into a new directory, record its exact commit, install locked dependencies without lifecycle scripts, build the SDK and source CLI, then define the shell command against that checkout. A clone or skill installation alone does not install the executable. Do not alter an existing active checkout to satisfy this path.
 

package/skills/slopcamera/references/music-video.md

@@ -1,5 +1,10 @@
 # Render a music video from an authored scene
 
+Choose the intended film or animation language with
+[visual style direction](visual-style-direction.md) before authoring the scene.
+The example dimensions below are an iteration setting; select native delivery
+dimensions explicitly for the finished movie.
+
 Use this workflow for a complete HTML or Three.js music video with a local track,
 including dancing characters, changing landscapes, and musical light accents.
 The export retains the source and creates an ordinary editable project with

package/skills/slopcamera/references/native-studio.md

@@ -1,5 +1,9 @@
 # Direct a native film studio
 
+For historical footage, physical materials, or a distinctive animation family,
+establish [visual style direction](visual-style-direction.md) before choosing
+lighting, camera, frame cadence, and finishing.
+
 These commands require current source; check [installation](install.md) and `slopcamera help studio`.
 
 Use `slopcamera studio` when the deliverable needs native Blender, CadQuery or Manim authoring. Keep the native source, caches and exact settings alongside review frames and final clips. This complements the existing Three/Spark scene and ordinary video-project workflows.

package/skills/slopcamera/references/patent-drawings.md

@@ -1,10 +1,10 @@
 # Prepare patent-style drawing sheets
 
 Use `diagram sheets` for local monochrome vector drawing sheets with physical
-page dimensions. These commands require current source; do not assume a
-previously published archive includes them. Confirm `slopcamera help diagram`
-lists the commands. If absent, follow [installation](install.md) for an explicit
-source build rather than inventing a flag on ordinary diagram rendering.
+page dimensions. The commands ship in Slopcamera v3.3.5 and later. Confirm
+`slopcamera help diagram` lists them. If they are absent, the installed CLI is
+older: follow [installation](install.md) to update it rather than inventing a
+flag on ordinary diagram rendering.
 
 ## Preserve the authored content
 

package/skills/slopcamera/references/support.md

@@ -6,7 +6,7 @@ At an eligible human closeout, call `slopcamera support offer --json` once. Trea
 around disabled state. Discovery on stderr is guidance, not an invitation claim.
 
 For `kind:offer`, optionally say once: “You can support continued development of
-Slopcamera’s local visual tools.” Use the returned clean support link. This product
+Slopcamera’s local media tools.” Use the returned clean support link. This product
 has no updates mailing list; do not collect an email or imply a subscription unlocks
 features. Opening a link grants no authority to pay or create an account.
 

package/skills/slopcamera/references/vectorization.md

@@ -9,7 +9,9 @@ slopcamera image vectorize path/to/input.png --output path/to/input.svg --json
 
 - Do not provide a credential for vectorization. The conversion keeps image bytes local; the initial tool provisioning can download a verified VTracer release.
 - Do not redraw, relabel, crop, recolor, or simplify the subject beyond the
-  trace mechanics the command measures.
+  trace mechanics the command measures. Brand illustrations are the exception:
+  they are normalized to the shared line language in
+  [brand illustrations](brand-illustrations.md), not preserved at fidelity.
 - Use `--duotone '#primary,#secondary'` only when the user explicitly asks for
   that two-color adaptation.
 - Preserve the emitted receipt in task output or an adjacent provenance record

package/skills/slopcamera/references/visual-style-direction.md

@@ -0,0 +1,143 @@
+# Direct a visual style
+
+Use this reference when a scene needs a recognizable film or animation language,
+or when producing a portfolio of distinct styles. The current-source `style`
+catalog records reusable production decisions:
+
+```sh
+slopcamera style list --json
+slopcamera style show silent-actuality --json
+slopcamera style show theatrical-cel --json
+```
+
+These profiles are art direction. They do not automatically transform an image,
+replace a scene's materials, synthesize character animation, or certify historical
+accuracy. Apply them through an appropriate renderer. The seven `html catalog`
+profiles describe rendering tools; visual styles are a separate choice.
+
+## Establish the picture before the finish
+
+Write a short direction for the specific subject: setting and period, dominant
+silhouette, foreground and distance planes, light source, camera position, one
+readable action, and intended emotional rhythm. Keep the user's literal facts and
+selected identity. Where invention is authorized, give the subject a concrete
+setting and action instead of filling the prompt with quality adjectives.
+
+Select the renderer that can produce the needed evidence:
+
+| Picture requirement | Useful route |
+| --- | --- |
+| Volumetric light, physical material, contact, perspective, architecture | Blender studio or Three scene |
+| Drawn contours, held drawings, flat shadow shapes, graphic composition | Canvas/p5/Two HTML scene |
+| Correct mathematical construction with deliberate teaching beats | Manim or checked deterministic Canvas geometry |
+| Photographic or painted source assets that procedural geometry cannot supply | Authorized Gateway image generation, then reviewed composition |
+| Photochemical finishing of an already convincing source | Existing ordered video-look effects, with restrained diffusion, grain, and vignette |
+
+A photo-textured background does not make moving geometry photorealistic. A
+Canvas material study does not demonstrate a physical clay simulation. Name the
+actual method in a portfolio caption.
+
+For creative delivery without specified dimensions, default to a native
+3840×2160 master or an equivalent frame in the intended aspect ratio when the
+renderer and source support it. Use explicit 1920×1080 previews for iteration.
+Respect source-image dimensions and a provider's actual output limits; state a
+lower native delivery clearly. Do not upscale a draft and call it native
+high-resolution detail. A 4:3 silent-film composition and a low-resolution pixel
+grid are deliberate exceptions to widescreen layout, not lower quality targets.
+
+## Historical film
+
+Choose a period and acquisition process before choosing damage. Early silent
+actuality, a 1940s studio negative, postwar 16mm documentary, and a 1970s home movie
+have different framing, motion, lighting, tonal latitude, and surface texture.
+Use a dated primary reference for specific history and label uncertain geometry
+or staging as interpretation. A source link beside an invented measurement does
+not verify that measurement.
+
+For an early street actuality:
+
+- Research street furniture, transport, overhead utilities, paving, clothing,
+  signage, and the sightline. Keep source-supported facts separate from estimated
+  dimensions and compressed geography. Inspect generated images for contradictions
+  even when the prompt explicitly forbids them. Do not add a familiar landmark or lamp
+  merely because it exists there today.
+- Stage a useful street depth: irregular storefront bays, window recesses,
+  cornices, varied reflectance, believable kerbs, contact shadows, and atmospheric
+  distance. Keep buildings from reading as identical extruded blocks.
+- Put the camera on the actual support, such as a cable car at a plausible human
+  height and speed. Keep distant geography coherent as it moves. Avoid a drone
+  orbit, wide modern lens, skyline reveal, or floating gimbal unless the reference
+  supports it.
+- Set capture cadence and playback cadence separately. A selected 16 fps
+  reconstruction is an artistic timing choice, not proof of the source camera's
+  exact crank speed. Hold drawings or photographed exposures deliberately; do
+  not interpolate every style to 60 fps.
+- Plan the monochrome values in the source. Early film spectral response cannot
+  be recovered by desaturating arbitrary modern colors. Make roofs, skin,
+  clothing, sky, and street separate in luminance; tune this against the reference.
+- Apply gentle highlight compression, optical softness, and restrained vignette
+  after the source works. Keep frame-varying grain fine enough to preserve facade
+  detail. Inspect wet surfaces, grain and stroke widths at final pixel size. Separate
+  visible sky brightness from ambient fill when a bright sky washes out contact
+  shadows; inspect material mapping on every visible face orientation.
+  Gate weave belongs to the entire image, not separate jittering objects.
+  Dirt and scratches are optional print wear, not the definition of old film.
+
+Distinguish capture, print, restoration, and scan. A clean high-resolution scan
+can retain fine texture and stable framing. Do not add heavy dust, crushed blacks,
+sepia, random flicker, letterbox bars, or a fake timecode by habit. Keep invented
+archival-looking footage labeled as an authored interpretation. A photographic
+plate with optical movement is a still-image study, not evidence of independently
+animated people, vehicles, or a historical event.
+
+The Market Street example includes a source manifest and an explicitly compressed
+street plan. Consult the Library of Congress film and SFMTA history linked in the
+example before changing its dates or street equipment.
+
+## Animation families
+
+An influence such as *Akira* or 3Blue1Brown is useful reference language. Translate
+it into mechanisms and create original subjects and compositions. The catalog
+uses generic technique names so the direction is reusable beyond one reference.
+
+| Family | Construct the image and movement | Reject at review |
+| --- | --- | --- |
+| Theatrical cel | Painted depth planes; precise perspective; selective contour weight; solid shadow shapes; articulated motion; held poses and short accents; camera motion sampled independently | Glossy toy materials, uniform outline filters, idle motion everywhere |
+| Watercolor storybook | Broad pigment washes, warm paper, lost edges, a focal patch of crisp ink, gentle overlapping action | A blur over vector art, evenly noisy outlines, identical texture on every object |
+| Ink sketch | Confident primary contours, sparse searching lines, directional hatching, selected line boil tied to held drawings | Unstable silhouettes, crosshatching through highlights, whole-scene wobble |
+| Rubber hose | Strong readable silhouette, curved limbs, arcs, anticipation, squash and stretch with volume awareness, clear contact | Random sine waves on every joint, sliding feet, impossible eye direction |
+| Midcentury limited | Asymmetric geometry, restrained spot palette, texture fields, selective held animation, strong negative space | Full constant motion, muddy colors, ornamental texture obscuring the subject |
+| Cut paper | Distinct cut contours, shallow layer separation, directional edge shadows, articulated pivots and reveal order | Universal blur, deep plastic bevels, disconnected shadows |
+| Stop-motion clay | Fingerprint-scale surface, soft contact, slightly imperfect construction, intentionally stepped poses and weighted settling | A smooth wobbling sphere described as physical stop motion, glossy uniform skin |
+| Pixel art | Explicit logical grid, limited palette, clustered shading, authored sprite poses, integer-aligned travel, nearest-neighbor integer scale | Blurred scaling, subpixel edges, unrestricted gradients, every object at a different pixel size |
+| Rotoscope | Gesture and weight from coherent anatomy, restrained contour drift, distinct held silhouettes, consistent volume | Unmotivated jitter, bendy joints, inconsistent limb lengths |
+| Engraving | Hatching follows form; line density describes value; leave highlights open; compose large black and white masses | Uniform screen-space stripe overlays or crushed unreadable detail |
+| Mathematical explanation | State one claim, assign stable colors, construct the relation, pause at its implication, keep notation readable | Decorative formulas, mismatched axes, moving labels, false equality, a waveform inconsistent with its generators |
+| Isometric design | Stable projection, consistent edge families, hierarchy through material and spacing, staged exploded movement | Conflicting vanishing points, changing scale, unrelated decorative labels |
+| Clean motion | A small set of shapes, optical alignment, purposeful easing, pauses, consistent transformation rules | Continuous bouncing, gratuitous overshoot, generic orbiting particles |
+
+Build the action from anticipation, main movement, follow-through, and a readable
+hold when those suit the style. Offset secondary action; avoid identical phase on
+every leaf or limb. Derive state from absolute frame time, with seeded variation
+keyed to the object and exposure index. Keep stable construction randomness
+separate from intentionally changing drawing or grain randomness. Seeking frames
+in a different order must reproduce the same image.
+
+## Review and retain
+
+Review representative frames before a full render: establishing view, strongest
+action, transition, and ending. Then inspect the encoded movie for timing,
+flicker, texture shimmer, crop, strobing, text, and compression. Pixel-level
+sharpness does not establish coherent art direction. A large output dimension
+does not establish source detail.
+
+For a portfolio, select different subjects and compositions that reveal each
+technique. Include the editable sources, profile IDs, actual output dimensions,
+render receipts, and honest renderer labels. Separate reusable direction from
+scene-specific drawing code. A contact sheet supports visual comparison; it does
+not replace checking motion. Preserve an old result when it explains a change.
+
+Keep bounded render jobs and their original sources. Dense grain can make a
+lossless intermediate exceed the existing media bound, especially at 4K. Shorten
+the shot or split it into retained takes without weakening that boundary. Do not
+retry an ambiguous native or paid generation before inspecting its retained state.

package/src/cli.ts

@@ -38,6 +38,7 @@ import { installSkill, type SkillScope, type SkillTarget } from "./skill-install
 import { pathExists } from "./fs.js"
 import { checkDrawingFile, renderDrawingFile, starterDrawingSource } from "./drawing.js"
 import { SLOPCAMERA_VERSION } from "./version.js"
+import { createVisualStyleDirection, getVisualStyleProfile, VISUAL_STYLE_PROFILES } from "./visual-style.js"
 import { reportUsefulResult, type UsefulResultObserver } from "./support-completion.js"
 import { runProductSupportCommand, showProductSupportInvitation, standaloneSupportEnvironment } from "./support.js"
 
@@ -62,6 +63,8 @@ Usage:
   slopcamera image gallery <subject> --output-dir <directory> [--kind <${slopcameraGalleryKinds.join("|")}>]
     [--count <1-${slopcameraGalleryLimits.candidates}>] [--vary <axis[=v1,v2][;axis...]>] [--candidates <file.json>]
     [--model <provider/model>] [--cell <${slopcameraGalleryLimits.cellEdgeMin}-${slopcameraGalleryLimits.cellEdgeMax}>] [--tile|--no-tile] [--json]
+  slopcamera style list [--json]
+  slopcamera style show <id> [--json]
   slopcamera code search [query] [--limit <number>]
   slopcamera code execute <operation> --input <JSON>
   slopcamera mcp --root <workspace>
@@ -370,6 +373,31 @@ export async function main(
   args: readonly string[],
   dependencies: SlopcameraCliDependencies = {},
 ): Promise<void> {
+  if (args[0] === "style") {
+    const [action, ...words] = args.slice(1)
+    const json = words.at(-1) === "--json"
+    const positionals = json ? words.slice(0, -1) : words
+    const log = dependencies.log ?? console.log
+    if (action === "list" && positionals.length === 0) {
+      log(json
+        ? JSON.stringify({ schemaVersion: 1, styles: VISUAL_STYLE_PROFILES })
+        : ["Visual style foundations (authoring guidance; no effects are applied):",
+          ...VISUAL_STYLE_PROFILES.map(style => `${style.id}  ${style.name}\n  ${style.summary}`),
+          "Use slopcamera style show <id> for palette, cadence, materials, delivery, and review criteria."].join("\n"))
+      return
+    }
+    if (action === "show" && positionals.length === 1 && !positionals[0]!.startsWith("-")) {
+      let style
+      try {
+        style = getVisualStyleProfile(positionals[0])
+      } catch {
+        throw new Error(`Unknown visual style: ${JSON.stringify(positionals[0]!.slice(0, 128))}. Use slopcamera style list.`)
+      }
+      log(json ? JSON.stringify({ schemaVersion: 1, style }) : createVisualStyleDirection(style.id))
+      return
+    }
+    throw new Error("Use slopcamera style list [--json] or slopcamera style show <id> [--json].")
+  }
   if (args[0] === "diagram" && args[1] === "sheets") {
     await runDrawingSheets(args.slice(2), dependencies)
     return

package/src/code/public-operations.ts

@@ -304,6 +304,15 @@ export interface PortableSlopcameraOperationResultMap {
   readonly "slopcamera.image.vectorize": SlopcameraImageVectorizeOutput
 }
 
+/**
+ * The typed portable projection: four of the six `slopcameraOperationCodes`.
+ * `slopcamera.image.icon` and `slopcamera.image.gallery` stay outside it on
+ * purpose. Each runs several paid Gateway calls and can publish more than one
+ * artifact, which the single-output, single-dispatch contract model here
+ * cannot express; they remain reachable through `executeSlopcameraOperation`,
+ * `execute_slopcamera`, and the CLI. Public copy cites this count, and
+ * `scripts/check-copy.ts` fails when the copy and this list disagree.
+ */
 export const PORTABLE_SLOPCAMERA_OPERATION_KINDS = Object.freeze([
   "slopcamera.diagram.check",
   "slopcamera.diagram.render",

package/src/index.ts

@@ -171,6 +171,7 @@ export * from "./operations.js"
 export * from "./portable-capability-manifest.js"
 export * from "./scene-gallery.js"
 export * from "./studio/index.js"
+export * from "./visual-style.js"
 export * from "./workflow.js"
 export type * from "./types.js"
 export type {

package/src/mcp/tools.ts

@@ -158,6 +158,10 @@ function deepFreeze<T>(value: T): T {
   return Object.freeze(value)
 }
 
+/**
+ * The complete MCP tool registry. Public copy cites its size, its scene-tool
+ * count, and its names; `scripts/check-copy.ts` fails when they disagree.
+ */
 export const slopcameraMcpTools: readonly McpToolDefinition[] = deepFreeze([
   {
     name: "check_diagram",

package/src/operations.ts

@@ -38,6 +38,10 @@ import {
   type HostResourceLease,
 } from "./host-resources.js"
 
+/**
+ * The fixed semantic operation registry. Public copy cites its size and names;
+ * `scripts/check-copy.ts` fails when README, docs, or skills disagree with it.
+ */
 export const slopcameraOperationCodes = [
   "slopcamera.diagram.check",
   "slopcamera.diagram.render",

package/src/support.ts

@@ -7,7 +7,7 @@ import {
 export const supportProfile = {
   id: "slopcamera",
   name: "Slopcamera",
-  valueProposition: "Support ongoing development of local visual tools for agents.",
+  valueProposition: "Support ongoing development of local media tools for agents.",
   updates: false,
 } as const
 

package/src/vectorize/command.ts

@@ -16,7 +16,12 @@ const MAX_COMMAND_OUTPUT_BYTES = 64 * 1_024
 const PRIMARY_OUTPUT_READ_BYTES = 64 * 1_024
 const PRIMARY_OUTPUT_POLL_MS = 2
 const PRIMARY_OUTPUT_EXIT_EMPTY_POLLS = 2
-const TERMINATION_GRACE_MS = 50
+/**
+ * Grace between SIGTERM and SIGKILL when a bounded command overruns. Callers
+ * that layer their own wall-clock budget above a bounded command must reserve
+ * at least this much, plus scheduling latency, so cleanup finishes inside it.
+ */
+export const TERMINATION_GRACE_MS = 50
 const HARD_KILL_WAIT_MS = 500
 const MAX_INHERITED_FILE_DESCRIPTORS = 16