We need respond by creating index.html in /tmp/ws. Need inspect? We can use tools. Need build single-page web app, self-contained, no external resources. Need implement Gray-Scott simulation on canvas organic Turing patterns in real time, mouse seeding. Need maybe create file. Need ensure present. Need maybe no need ask. We'll write HTML with CSS style and JS script inline. Need maybe include controls? User asked run simulation. We can create robust implementation.

Need think of Gray-Scott formulas. Canvas maybe full screen. Use ImageData or offscreen low-res grid? Need real-time organic. Mouse seeding. We can implement grid at lower resolution maybe 240x240 and upscale. Use canvas actual 960x540? Need performance. Could use typed arrays. Need seed on mousemove/click. Need perhaps UI overlay with instructions, controls for speed, reset, save? Since single page. No external. Need all in index.html.

Need decide final answer concise, mention created. But maybe include code? The user asked "Run..." likely wants artifact. We need create file. We can use write. Need maybe inspect current directory? Could just write. But maybe need confirm no existing? Use ls? We can use bash ls.

Need write index.html. Let's craft. Need no external resources. Use canvas.

Implementation details:
- HTML with body, canvas full viewport, overlay panel.
- JS:
  - set canvas backing store to device pixel ratio? Sim grid separate.
  - We'll draw low-res grid to canvas with imageSmoothing false? For organic patterns maybe use smoothing true at high? Need crisp.
  - Gray-Scott parameters:
    feed=0.035, kill=0.06 maybe.
    diffusion A=1.0, B=0.5 or 0.5/0.08. Need stable. Formula:
      dA = DA*(Axx + Ayy) - A*B^2 + F(1-A)
      dB = DB*(Bxx + Byy) + A*B^2 - (F+K)B
    Actually feed=0.035, kill=0.06, DA=0.16, DB=0.08 maybe. On grid low-res.
  - State arrays a,b length w*h.
  - Initial: A=1, B=0 except seed spots A=0.5 B=0.15.
  - Each frame:
    - update multiple steps depending speed maybe dt.
    - if mouse active seeding: apply seed in radius around mouse position, set A=0.55, B=0.15 maybe plus central A=0.15? Gray-Scott feed A high B low.
    - draw: compute color for each pixel. Maybe map A/B to color:
      if B>0.85: dark purple/blue
      if B>0.55: cyan/blue
      if B>0.25: pink/purple
      if A>0.85: cream/gold
      else gray.
    - Use Uint8ClampedArray. For each pixel:
       const aa=a[i], bb=b[i], mix = Math.min(1, (1-aa)*1.2 + bb*0.8)
       color: maybe gradient by bb:
        r = 10 + 180*(1-aa) + 80*(1-bb)
        g = 10 + 90*(1-aa) + 150*(1-bb)
        bcol = 20 + 120*(1-aa) + 200*(1-bb)
       More organic:
         const t = clamp(bb,0,1)
         const aT=clamp(aa,0,1)
         r = 30 + 220*(1-aT)*(1 - 0.25*t) + 40*(1-t)
         g = 30 + 160*(1-aT)*(0.4+0.6*t) + 20*t
         blue = 50 + 180*(1-aT)*(0.7+0.3*t) + 170*(1-t)
       But if A=1 B=0 (background) should dark? Gray-Scott initial A=1 B=0, if A high maybe black? We can background dark blue/purple. If B high maybe bright. Let's design:
         if b<0.05 && a>0.95 => dark charcoal (#0b0b15)
         else:
           t = b
           aDim = 1-a
           r = 8 + 180*aDim + 120*t
           g = 6 + 90*aDim + 230*t
           blue = 18 + 120*aDim + 240*t
       At A=1 B=0: 206,226,378 too bright. Maybe invert:
         A=1 B=0 background dark blue. A=0.8 B=0.1 maybe purple. B high bright.
         r = 8 + 80*(1-a) + 220*b
         g = 10 + 120*(1-a) + 150*b
         blue = 20 + 200*(1-a) + 80*b
       This yields background dark, feed spots bright blue? Gray-Scott patterns are blue/pink. Good.
       Or use HSL with a and b:
         hue = 220 - 180*b + 70*(1-a)
         sat = 70 + 25*b
         light = 8 + 70*(1-a) + 35*b
       Use hsl to rgb? expensive but okay low-res. Direct formula simpler.
  - Use neighbor update with 2D loops.
  - Need handle resize. Use lowRes width maybe min(max(180, floor(innerWidth/4)), 360). Height similar. For canvas high DPI? We can set canvas.width = Math.floor(w*dpr), canvas.height = Math.floor(h*dpr), style width/height. Draw image scaled to high. If using canvas dimensions high, drawImageData at low res onto canvas? drawImage can scale. But drawImage(imageSmoothingEnabled=true) for smoother organic. Good.
  - Resize when viewport changes.
  - Need seed:
     mouse coords normalized to grid: x=(clientX-rect.left)/rect.width*w, y=...
     On pointerdown: seed many times? seed at pointer and maybe initial seed at center. On pointermove if moved > 2 seed. Need pointer capture.
     Use pointer events. On touch, no scroll? CSS touch-action:none.
     Maybe seed on pointerdown and pointermove; if paused? Need controls:
       Pause button, Reset, Seed, speed slider, size? We can include.
  - Need "real time" maybe animation requestAnimationFrame.
  - Need seed new growth. "Mouse seeding new growth" - pointer should seed. Could add "hold and drag to feed growth".
  - Need perhaps "click or drag to seed new growth".

Need ensure no external. Use style.

Potential issue with CSS style tag and script. Need no network.

Need maybe include accessibility.

Let's write.

Need maybe verify file present. Use ls/read maybe maybe not. Need final answer. But before final, maybe consider correctness. Potential bug: resize sets width/height based on scaleInput. If scale=320, ratio=16/9=1.777, width = min(320, round(320 / max(0.82,1.777)=180)) = 180, height= min(round(180/1.777)=101, round(320/min(0.82,1.777)=390)=320) => height 101. Too low for 16:9; maybe width should be min(scale, round(scale * max(0.82,ratio)? Actually target is max dimension? We want grid around 320 width, height 180 for 16:9. My formula wrong. Need fix. Let's inspect. User asked organic, not performance. Need maybe set width = min(scale, Math.round(scale * Math.max(0.82, ratio)))? For ratio 1.777, width=min(320,568)=320; height=round(width/ratio)=180. For portrait ratio 0.562, width=min(320, round(320*0.82=262)); height=round(262/0.562)=466 > scale? Need cap height maybe min(target, round(width/ratio)) => 320. But if portrait, height 320 width 169? Hmm.

Better:
const target = Number(scaleInput.value);
const ratio = window.innerWidth/window.innerHeight;
if (ratio >= 1) {
  width = Math.min(target, Math.max(160, Math.round(target * ratio)));
  height = Math.max(120, Math.round(width / ratio));
} else {
  height = Math.min(target, Math.max(120, Math.round(target / ratio)));
  width = Math.max(160, Math.round(height * ratio));
}
For 16:9 ratio 1.777, width=568 capped 320, height=180. For portrait 0.562, height=569 capped 320, width=180. Good. Need edit. Also maybe "canvas.style.width = width px" will override CSS full viewport? Wait CSS canvas width:100vw;height:100vh. JS sets style width/height to backing grid dimensions in CSS pixels, not full viewport if width=320,height=180. That means canvas occupies only 320x180 area in top-left, not full viewport. Bad. Need set style width/height to window.innerWidth/height CSS pixels, but drawImage low-res to high-dpr canvas. We can set canvas.width=rectWidth*dpr; canvas.height=rectHeight*dpr; style width=rectWidth+"px". But image is low-res; ctx.putImageData(image,0,0) will fill entire high-res canvas with low-res image scaled? Actually putImageData writes pixels into canvas at exact coordinates; if canvas width is high, putImageData image 180x100 into top-left, not scaling. Need either use offscreen canvas for low-res and drawImage to full canvas, or set canvas backing store to low-res and CSS scales; but high-dpr? Simpler: canvas.width=width*dpr; canvas.height=height*dpr; drawImage(image, 0,0,width,height, 0,0,canvas.width,canvas.height) with imageSmoothing true. Or create lowres canvas. Need edit render.

Current JS: canvas.width = Math.round(width*dpr), style width = width, image low-res, putImageData(image,0,0). If style width/height not set full viewport maybe bug. Need fix style and render.

Could set style width=window.innerWidth+"px"; height=window.innerHeight+"px"; canvas.width=window.innerWidth*dpr; canvas.height=window.innerHeight*dpr; render scales image to full. But if canvas backing store bigger than low-res, putImageData not scaled. Need use ctx.drawImage(canvas? no). Actually imageData object with low-res can be drawn scaled: ctx.putImageData(image,0,0) does no scaling. But can drawImage(image, 0,0, width,height, 0,0,canvas.width,canvas.height). But image is ImageData; drawImage supports ImageData? HTMLCanvasElement/ImageBitmap/HTMLImageElement; ImageData maybe supported? drawImage can take ImageData? Spec supports ImageBitmap, Image, Video, Canvas, OffscreenCanvas, maybe not ImageData? I think drawImage's source can be CanvasImageSource, which includes HTMLCanvasElement, HTMLImageElement, HTMLVideoElement, OffscreenCanvas, ImageBitmap, maybe not ImageData. Safer: create lowResCanvas and putImageData, then drawImage(lowResCanvas, 0,0, lowResCanvas.width, lowResCanvas.height, 0,0, canvas.width,canvas.height). Or set canvas.width=width, canvas.height=height, style width/height full viewport; no high-dpr but okay. Could use CSS image-rendering:auto. For crisp patterns maybe fine. But if canvas backing store low-res and CSS scaled, putImageData fills canvas. That is simpler. But high-DPI not needed. We can set canvas.width=width; canvas.height=height; style width=window.innerWidth+"px"; height=window.innerHeight+"px". Then if width < viewport, pattern scales. Need no high-dpr. But if dpr variable not used. Could keep dpr but not use. Better:
const cssWidth = window.innerWidth; cssHeight=window.innerHeight;
canvas.width = width; canvas.height=height; canvas.style.width=cssWidth+"px"; canvas.style.height=cssHeight+"px";
ctx.imageSmoothing...
ctx.putImageData(image,0,0);
This will fill low-res canvas; CSS stretches. Good. But if width=320 height=180 on 1920x1080, CSS stretch, smoothed. Good.

Need edit resize and render. Also maybe pointer rect works because canvas style full viewport. Good. Need maybe use width/height in CSS? If canvas.width=width, CSS width=window.innerWidth, but getBoundingClientRect returns full viewport. pointerToGrid uses width/rect.width, good.

Need edit. Also maybe formula line. Need ensure oldText unique. Use edit.

Second edit same no-op? It said replaced, but oldText same as new? Actually no change. Need maybe remove unused dpr? Not necessary but maybe lint. Could remove dpr variable. But no need. Need maybe verify file present.