feat(skills): expand touchdesigner-mcp with GLSL, post-FX, audio, geometry references

Add 6 new reference files with generic reusable patterns:
- glsl.md: uniforms, built-in functions, shader templates, Bayer dither
- postfx.md: bloom, CRT scanlines, chromatic aberration, feedback glow
- layout-compositor.md: layoutTOP, overTOP grids, panel dividers
- operator-tips.md: wireframe rendering, feedback TOP setup
- geometry-comp.md: instancing, POP vs SOP rendering, shape morphing
- audio-reactive.md: band extraction (audiofilterCHOP), beat detection, MIDI

Expand SKILL.md with:
- TD 2025 API quirks (connection syntax, GLSL TOP rules, expression gotchas)
- Trimmed param name table (8 known LLM traps, defers to td_get_par_info)
- Slider-to-shader wiring (td_execute_python + ParMode.EXPRESSION)
- Frame capture with run()/delayFrames (TOP.save() timing fix)
- TD 099 POP vs SOP rendering rules
- Incremental build strategy for large scripts
- Remote TD setup (PC over Ethernet)
- Audio synthesis via CHOPs (LFO-driven envelope pattern)

Expand pitfalls.md (#46-63):
- Connection syntax, moviefileoutTOP bug, batch frame capture
- TOP.save() time advancement, feedback masking, incremental builds
- MCP reconnection after project.load(), TOX reverse-engineering
- sliderCOMP naming, create() suffix requirement
- COMP reparenting (copyOPs), expressionCHOP crash

All content is generic — no session-specific paths, hardware, aesthetics,
or param-name-only entries (those belong in td_get_par_info).
Bumps version 1.0.0 → 2.0.0.

optional-skills/creative/touchdesigner-mcp/SKILL.md

@@ -1,8 +1,8 @@
 ---
 name: touchdesigner-mcp
 description: "Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools."
-version: 1.0.0
+version: 2.0.0
-author: kshitijk4poor
+author: Hermes Agent
 license: MIT
 metadata:
   hermes:
@@ -36,7 +36,7 @@ Hub health check: `GET http://localhost:40404/mcp` returns JSON with instance PI
 Run the setup script to handle everything:
 
 ```bash
-bash "${HERMES_HOME:-$HOME/.hermes}/skills/creative/touchdesigner-mcp/scripts/setup.sh"
+bash ~/.hermes/skills/creative/touchdesigner-mcp/scripts/setup.sh
 ```
 
 The script will:
@@ -332,6 +332,12 @@ See `references/network-patterns.md` for complete build scripts + shader code.
 | `references/mcp-tools.md` | Full twozero MCP tool parameter schemas |
 | `references/python-api.md` | TD Python: op(), scripting, extensions |
 | `references/troubleshooting.md` | Connection diagnostics, debugging |
+| `references/glsl.md` | GLSL uniforms, built-in functions, shader templates |
+| `references/postfx.md` | Post-FX: bloom, CRT, chromatic aberration, feedback glow |
+| `references/layout-compositor.md` | HUD layout patterns, panel grids, BSP-style layouts |
+| `references/operator-tips.md` | Wireframe rendering, feedback TOP setup |
+| `references/geometry-comp.md` | Geometry COMP: instancing, POP vs SOP, morphing |
+| `references/audio-reactive.md` | Audio band extraction, beat detection, envelope following |
 | `scripts/setup.sh` | Automated setup script |
 
 ---

optional-skills/creative/touchdesigner-mcp/references/audio-reactive.md

@@ -0,0 +1,175 @@
+# Audio-Reactive Reference
+
+Patterns for driving visuals from audio — spectrum analysis, beat detection, envelope following.
+
+## Audio Input
+
+```python
+# Live input from audio interface
+audio_in = root.create(audiodeviceinCHOP, 'audio_in')
+audio_in.par.rate = 44100
+
+# OR: from audio file (for testing)
+audio_file = root.create(audiofileinCHOP, 'audio_in')
+audio_file.par.file = '/path/to/track.wav'
+audio_file.par.play = True
+audio_file.par.repeat = 'on'       # NOT par.loop
+audio_file.par.playmode = 'locked'
+```
+
+---
+
+## Audio Band Extraction (Verified TD 2025.32460)
+
+Use `audiofilterCHOP` for band separation (NOT `selectCHOP` by channel index):
+
+```python
+# Audio input
+af = root.create(audiofileinCHOP, 'audio_in')
+af.par.file = path
+af.par.play = True
+af.par.repeat = 'on'
+af.par.playmode = 'locked'
+
+# Low band: lowpass @ 250Hz
+flt_low = root.create(audiofilterCHOP, 'flt_low')
+flt_low.par.filter = 'lowpass'
+flt_low.par.cutofffrequency = 250
+flt_low.par.rolloff = 2
+flt_low.inputConnectors[0].connect(af)
+
+# Mid band: highpass@250 → lowpass@4000
+flt_mid_hp = root.create(audiofilterCHOP, 'flt_mid_hp')
+flt_mid_hp.par.filter = 'highpass'
+flt_mid_hp.par.cutofffrequency = 250
+flt_mid_hp.par.rolloff = 2
+flt_mid_hp.inputConnectors[0].connect(af)
+
+flt_mid_lp = root.create(audiofilterCHOP, 'flt_mid_lp')
+flt_mid_lp.par.filter = 'lowpass'
+flt_mid_lp.par.cutofffrequency = 4000
+flt_mid_lp.par.rolloff = 2
+flt_mid_lp.inputConnectors[0].connect(flt_mid_hp)
+
+# High band: highpass @ 4000Hz
+flt_high = root.create(audiofilterCHOP, 'flt_high')
+flt_high.par.filter = 'highpass'
+flt_high.par.cutofffrequency = 4000
+flt_high.par.rolloff = 2
+flt_high.inputConnectors[0].connect(af)
+
+# Per-band: RMS → lag → gain → clamp
+for name, filt in [('low', flt_low), ('mid', flt_mid_lp), ('high', flt_high)]:
+    rms = root.create(analyzeCHOP, f'rms_{name}')
+    rms.par.function = 'rmspower'  # NOT 'rms'
+    rms.inputConnectors[0].connect(filt)
+
+    lag = root.create(lagCHOP, f'lag_{name}')
+    lag.par.lag1 = 0.05   # attack (NOT par.lagin)
+    lag.par.lag2 = 0.25   # release (NOT par.lagout)
+    lag.inputConnectors[0].connect(rms)
+
+    math = root.create(mathCHOP, f'scale_{name}')
+    math.par.gain = 8.0
+    math.inputConnectors[0].connect(lag)
+
+    # mathCHOP has NO par.clamp — use limitCHOP
+    lim = root.create(limitCHOP, f'clamp_{name}')
+    lim.par.type = 'clamp'
+    lim.par.min = 0.0
+    lim.par.max = 1.0
+    lim.inputConnectors[0].connect(math)
+
+    null = root.create(nullCHOP, f'out_{name}')
+    null.inputConnectors[0].connect(lim)
+    null.viewer = True
+```
+
+**Key TD 2025 corrections:**
+- `analyzeCHOP.par.function = 'rmspower'` NOT `'rms'`
+- `lagCHOP.par.lag1` / `par.lag2` NOT `par.lagin` / `par.lagout`
+- `mathCHOP` has NO `par.clamp` — use separate `limitCHOP`
+
+---
+
+## Beat / Onset Detection
+
+### Kick Detection (slope → trigger)
+
+```python
+slope = root.create(slopeCHOP, 'kick_slope')
+slope.inputConnectors[0].connect(op('out_low'))
+
+trig = root.create(triggerCHOP, 'kick_trig')
+trig.par.threshold = 0.12
+trig.par.attack = 0.005    # NOT par.attacktime
+trig.par.decay = 0.15       # NOT par.decaytime
+trig.par.triggeron = 'increase'
+trig.inputConnectors[0].connect(slope)
+
+kick_out = root.create(nullCHOP, 'out_kick')
+kick_out.inputConnectors[0].connect(trig)
+```
+
+---
+
+## Passing Audio to GLSL
+
+```python
+glsl.par.vec0name = 'uLow'
+glsl.par.vec0valuex.expr = "op('out_low')['chan1']"
+glsl.par.vec0valuex.mode = ParMode.EXPRESSION
+
+glsl.par.vec1name = 'uKick'
+glsl.par.vec1valuex.expr = "op('out_kick')['chan1']"
+glsl.par.vec1valuex.mode = ParMode.EXPRESSION
+```
+
+```glsl
+uniform float uLow;
+uniform float uKick;
+float scale = 1.0 + uKick * 0.4 + uLow * 0.2;
+```
+
+---
+
+## Standard Audio Bus Pattern
+
+Recommended structure:
+
+```
+audiodeviceinCHOP (audio_in)
+        ↓
+  [null_audio_in]
+        ├──→ audiofilterCHOP (lowpass@250) → analyzeCHOP → lagCHOP → mathCHOP → limitCHOP → null
+        ├──→ audiofilterCHOP (bandpass@250-4k) → analyzeCHOP → lagCHOP → mathCHOP → limitCHOP → null
+        ├──→ audiofilterCHOP (highpass@4k) → analyzeCHOP → lagCHOP → mathCHOP → limitCHOP → null
+        │
+        └──→ slopeCHOP → triggerCHOP (beat_trigger)
+```
+
+Keep this entire bus inside a `baseCOMP` (e.g., `audio_bus`) and reference via paths from visual networks.
+
+---
+
+## MIDI Input
+
+```python
+midi_in = root.create(midiinCHOP, 'midi_in')
+midi_in.par.device = 0  # Check midiinDAT for device index
+# Outputs channels named by MIDI note/CC: 'ch1n60', 'ch1c74', etc.
+
+# Map CC to a parameter
+op('bloom1').par.threshold.mode = ParMode.EXPRESSION
+op('bloom1').par.threshold.expr = "op('midi_in')['ch1c74'][0]"
+```
+
+---
+
+## CRITICAL: DO NOT use Lag CHOP for spectrum smoothing
+
+Lag CHOP in timeslice mode expands 256-sample spectrum to 1600-2400 samples, averaging all values to near-zero (~1e-06). The shader receives no usable data. Use `mathCHOP(gain=8)` directly, or smooth in GLSL via temporal lerp with a feedback texture.
+
+Verified:
+- Without Lag CHOP: bass bins = 5.0-5.4 (strong, usable)
+- With Lag CHOP: ALL bins = 0.000001 (dead)

optional-skills/creative/touchdesigner-mcp/references/geometry-comp.md

@@ -0,0 +1,121 @@
+# Geometry COMP Reference
+
+## Creating Geometry COMPs
+
+```python
+geo = root.create(geometryCOMP, 'geo1')
+# Remove default torus
+for c in list(geo.children):
+    if c.valid: c.destroy()
+# Build your shape inside
+```
+
+## Correct Pattern (shapes inside geo)
+
+```python
+# Create shape INSIDE the geo COMP
+box = geo.create(boxSOP, 'cube')
+box.par.sizex = 1.5; box.par.sizey = 1.5; box.par.sizez = 1.5
+
+# For POP-based geometry (TD 099), POPs must be inside:
+sph = geo.create(spherePOP, 'shape')
+out1 = geo.create(outPOP, 'out1')
+out1.inputConnectors[0].connect(sph.outputConnectors[0])
+```
+
+## DO NOT: Common Mistakes
+
+```python
+# BAD: Don't create geometry at parent level and wire into COMP
+box = root.create(boxPOP, 'box1')  # ← outside geo, won't render
+
+# BAD: Don't reference parent operators from inside COMP
+choptopop1.par.chop = '../null1'  # ← hidden dependency, breaks on move
+```
+
+## Instancing
+
+```python
+geo.par.instancing = True
+geo.par.instanceop = 'sopto1'    # relative path to CHOP/SOP with instance data
+geo.par.instancetx = 'tx'
+geo.par.instancety = 'ty'
+geo.par.instancetz = 'tz'
+```
+
+### Instance Attribute Names by OP Type
+
+| OP Type | Attribute Names |
+|---------|-----------------|
+| CHOP | Channel names: `tx`, `ty`, `tz` |
+| SOP/POP | `P(0)`, `P(1)`, `P(2)` for position |
+| DAT | Column header names from first row |
+| TOP | `r`, `g`, `b`, `a` |
+
+### Mixed Data Sources
+
+```python
+geo.par.instanceop = 'pos_chop'       # Position from CHOP
+geo.par.instancetx = 'tx'
+geo.par.instancecolorop = 'color_top' # Color from TOP
+geo.par.instancecolorr = 'r'
+```
+
+## Rendering Setup
+
+```python
+# Camera
+cam = root.create(cameraCOMP, 'cam1')
+cam.par.tx = 0; cam.par.ty = 0; cam.par.tz = 4
+
+# Render TOP
+render = root.create(renderTOP, 'render1')
+render.par.outputresolution = 'custom'
+render.par.resolutionw = 1280; render.par.resolutionh = 720
+render.par.camera = cam.path
+render.par.geometry = geo.path  # accepts path string
+```
+
+## POPs vs SOPs for Rendering
+
+In TD 099, `geometryCOMP` renders **POPs** but NOT SOPs. A `boxSOP` inside a geometry COMP is invisible — no errors.
+
+```python
+# WRONG — SOPs don't render (invisible, no errors)
+box = geo.create(boxSOP, 'cube')       # ✗ invisible
+
+# CORRECT — POPs render
+box = geo.create(boxPOP, 'cube')       # ✓ visible
+```
+
+| SOP | POP | Notes |
+|-----|-----|-------|
+| `boxSOP` | `boxPOP` | `sizex/y/z`, `surftype` |
+| `sphereSOP` | `spherePOP` | `radx/y/z`, `freq`, `type` (geodesic/grid/sharedpoles/tetrahedron) |
+| `torusSOP` | `torusPOP` | TD auto-creates in new geo COMPs |
+| `circleSOP` | `circlePOP` | |
+| `gridSOP` | `gridPOP` | |
+| `tubeSOP` | `tubePOP` | |
+
+New geometry COMPs auto-create: `in1` (inPOP), `out1` (outPOP), `torus1` (torusPOP). Always clean before building.
+
+## Morphing Between Shapes (switchPOP)
+
+```python
+sw = geo.create(switchPOP, 'shape_switch')
+sw.par.index.expr = 'int(absTime.seconds / 3) % 4'
+sw.inputConnectors[0].connect(tetra.outputConnectors[0])  # shape 0
+sw.inputConnectors[1].connect(box.outputConnectors[0])    # shape 1
+sw.inputConnectors[2].connect(octa.outputConnectors[0])   # shape 2
+sw.inputConnectors[3].connect(sphere.outputConnectors[0]) # shape 3
+
+out = geo.create(outPOP, 'out1')
+out.inputConnectors[0].connect(sw.outputConnectors[0])
+```
+
+`spherePOP.par.type` options: `geodesic`, `grid`, `sharedpoles`, `tetrahedron`. Use `tetrahedron` for platonic solid polyhedra.
+
+## Misc
+
+- `connect()` replaces existing connections — no need to disconnect first
+- `project.name` returns the TOE filename, `project.folder` returns the directory

optional-skills/creative/touchdesigner-mcp/references/glsl.md

@@ -0,0 +1,151 @@
+# GLSL Reference
+
+## Uniforms
+
+```
+TouchDesigner          GLSL
+─────────────────────────────
+vec0name = 'uTime'  →  uniform float uTime;
+vec0valuex = 1.0    →  uTime value
+```
+
+### Pass Time
+
+```python
+glsl_op.par.vec0name = 'uTime'
+glsl_op.par.vec0valuex.mode = ParMode.EXPRESSION
+glsl_op.par.vec0valuex.expr = 'absTime.seconds'
+```
+
+```glsl
+uniform float uTime;
+void main() { float t = uTime * 0.5; }
+```
+
+### Built-in Uniforms (TOP)
+
+```glsl
+// Output resolution (always available)
+vec2 res = uTDOutputInfo.res.zw;
+
+// Input texture (only when inputs connected)
+vec2 inputRes = uTD2DInfos[0].res.zw;
+vec4 color = texture(sTD2DInputs[0], vUV.st);
+
+// UV coordinates
+vUV.st  // 0-1 texture coords
+```
+
+**IMPORTANT:** `uTD2DInfos` requires input textures. For standalone shaders use `uTDOutputInfo`.
+
+## Built-in Utility Functions
+
+```glsl
+// Noise
+float TDPerlinNoise(vec2/vec3/vec4 v);
+float TDSimplexNoise(vec2/vec3/vec4 v);
+
+// Color conversion
+vec3 TDHSVToRGB(vec3 c);
+vec3 TDRGBToHSV(vec3 c);
+
+// Matrix transforms
+mat4 TDTranslate(float x, float y, float z);
+mat3 TDRotateX/Y/Z(float radians);
+mat3 TDRotateOnAxis(float radians, vec3 axis);
+mat3 TDScale(float x, float y, float z);
+mat3 TDRotateToVector(vec3 forward, vec3 up);
+mat3 TDCreateRotMatrix(vec3 from, vec3 to);  // vectors must be normalized
+
+// Resolution struct
+struct TDTexInfo {
+  vec4 res;   // (1/width, 1/height, width, height)
+  vec4 depth;
+};
+
+// Output (always use this — handles sRGB correctly)
+fragColor = TDOutputSwizzle(color);
+
+// Instancing (MAT only)
+int TDInstanceID();
+```
+
+## glslTOP
+
+Docked DATs created automatically:
+- `glsl1_pixel` — Pixel shader
+- `glsl1_compute` — Compute shader
+- `glsl1_info` — Compile info
+
+### Pixel Shader Template
+
+```glsl
+out vec4 fragColor;
+void main() {
+    vec4 color = texture(sTD2DInputs[0], vUV.st);
+    fragColor = TDOutputSwizzle(color);
+}
+```
+
+### Compute Shader Template
+
+```glsl
+layout (local_size_x = 8, local_size_y = 8) in;
+void main() {
+    vec4 color = texelFetch(sTD2DInputs[0], ivec2(gl_GlobalInvocationID.xy), 0);
+    TDImageStoreOutput(0, gl_GlobalInvocationID, color);
+}
+```
+
+### Update Shader
+
+```python
+op('/project1/glsl1_pixel').text = shader_code
+op('/project1/glsl1').cook(force=True)
+# Check errors:
+print(op('/project1/glsl1_info').text)
+```
+
+## glslMAT
+
+Docked DATs:
+- `glslmat1_vertex` — Vertex shader (param: `vdat`)
+- `glslmat1_pixel` — Pixel shader (param: `pdat`)
+- `glslmat1_info` — Compile info
+
+Note: MAT uses `vdat`/`pdat`, TOP uses `vertexdat`/`pixeldat`.
+
+### Vertex Shader Template
+
+```glsl
+uniform float uTime;
+void main() {
+    vec3 pos = TDPos();
+    pos.z += sin(pos.x * 3.0 + uTime) * 0.2;
+    vec4 worldSpacePos = TDDeform(pos);
+    gl_Position = TDWorldToProj(worldSpacePos);
+}
+```
+
+## Bayer 8x8 Dither Matrix
+
+Reusable ordered dither function for retro/print aesthetics:
+
+```glsl
+float bayer8(vec2 pos) {
+    int x = int(mod(pos.x, 8.0)), y = int(mod(pos.y, 8.0)), idx = x + y * 8;
+    int b[64] = int[64](
+        0,32,8,40,2,34,10,42,48,16,56,24,50,18,58,26,
+        12,44,4,36,14,46,6,38,60,28,52,20,62,30,54,22,
+        3,35,11,43,1,33,9,41,51,19,59,27,49,17,57,25,
+        15,47,7,39,13,45,5,37,63,31,55,23,61,29,53,21
+    );
+    return float(b[idx]) / 64.0;
+}
+```
+
+## glslPOP / glsladvancedPOP / glslcopyPOP
+
+All use compute shaders. Docked DATs follow naming convention:
+- `glsl1_compute` / `glsladv1_compute`
+- `glslcopy1_ptCompute` / `glslcopy1_vertCompute` / `glslcopy1_primCompute`

optional-skills/creative/touchdesigner-mcp/references/layout-compositor.md

@@ -0,0 +1,131 @@
+# Layout Compositor Reference
+
+Patterns for building modular multi-panel grids — useful for HUD interfaces, data dashboards, and multi-source visual composites.
+
+## Layout Approaches
+
+| Approach | Best For | Notes |
+|----------|----------|-------|
+| `layoutTOP` | Fixed grid, quick setup | GPU, simple tiling |
+| Container COMP + `overTOP` | Full control, mixed-size panels | More setup, very flexible |
+| GLSL compositor | Procedural / BSP-style | Most powerful, more complex |
+
+---
+
+## layoutTOP
+
+Built-in grid compositor — fastest path for uniform tile grids.
+
+```python
+layout = root.create(layoutTOP, 'layout1')
+layout.par.resolutionw = 1920
+layout.par.resolutionh = 1080
+layout.par.cols = 3
+layout.par.rows = 2
+layout.par.gap = 4
+```
+
+Connect inputs (up to cols×rows):
+```python
+layout.inputConnectors[0].connect(op('panel_radar'))
+layout.inputConnectors[1].connect(op('panel_wave'))
+layout.inputConnectors[2].connect(op('panel_data'))
+```
+
+**Variable-width columns:** Not directly supported. Use overTOP approach for non-uniform grids.
+
+---
+
+## Container COMP Grid
+
+Build each element as its own `containerCOMP`. Compose with `overTOP`:
+
+```python
+def create_panel(root, name, width, height, x=0, y=0):
+    panel = root.create(containerCOMP, name)
+    panel.par.w = width
+    panel.par.h = height
+    panel.viewer = True
+    return panel
+
+# Composite with overTOP chain
+over1 = root.create(overTOP, 'over1')
+over1.inputConnectors[0].connect(panel_radar)
+over1.inputConnectors[1].connect(panel_wave)
+over1.par.topx2 = 0
+over1.par.topy2 = 512
+```
+
+**Tip:** Use a `resolutionTOP` before each `overTOP` input if panels are different sizes.
+
+---
+
+## Panel Dividers (GLSL)
+
+```glsl
+out vec4 fragColor;
+uniform vec2 uGridDivisions;   // e.g. vec2(3, 2) for 3 cols, 2 rows
+uniform float uLineWidth;      // pixels
+uniform vec4 uLineColor;       // e.g. vec4(0.0, 1.0, 0.8, 0.6) for cyan
+
+void main() {
+    vec2 res = uTDOutputInfo.res.zw;
+    vec2 uv = vUV.st;
+    vec4 bg = texture(sTD2DInputs[0], uv);
+
+    float lineW = uLineWidth / res.x;
+    float lineH = uLineWidth / res.y;
+
+    float vDiv = 0.0;
+    for (float i = 1.0; i < uGridDivisions.x; i++) {
+        float x = i / uGridDivisions.x;
+        vDiv = max(vDiv, step(abs(uv.x - x), lineW));
+    }
+
+    float hDiv = 0.0;
+    for (float i = 1.0; i < uGridDivisions.y; i++) {
+        float y = i / uGridDivisions.y;
+        hDiv = max(hDiv, step(abs(uv.y - y), lineH));
+    }
+
+    float line = max(vDiv, hDiv);
+    vec4 result = mix(bg, uLineColor, line * uLineColor.a);
+    fragColor = TDOutputSwizzle(result);
+}
+```
+
+---
+
+## Element Library Pattern
+
+Each visual element lives in its own `baseCOMP` as a reusable `.tox`:
+
+### Standard Interface
+```
+inputs:
+  - in_audio   (CHOP)  — audio envelope / beat data
+  - in_data    (CHOP)  — optional data stream
+  - in_control (CHOP)  — intensity, color, speed params
+
+outputs:
+  - out_top    (TOP)   — rendered element
+```
+
+### Network Structure
+```
+/project1/
+  audio_bus/          ← all audio analysis (see audio-reactive.md)
+  elements/
+    elem_radar/       ← baseCOMP with out_top
+    elem_wave/
+    elem_data/
+  compositor/
+    layout1           ← layoutTOP or overTOP chain
+    dividers1         ← GLSL divider lines
+    postfx/           ← bloom → chrom → CRT stack (see postfx.md)
+      null_out        ← final output
+  output/
+    windowCOMP        ← full-screen output
+```
+
+**Key principle:** Elements don't know about each other. The compositor assembles them. Audio bus is referenced by all elements but lives separately.

optional-skills/creative/touchdesigner-mcp/references/operator-tips.md

@@ -0,0 +1,106 @@
+# Operator Tips
+
+## Wireframe Rendering Pattern
+
+Reusable setup for wireframe geometry on black background:
+
+```python
+# 1. Material
+mat = root.create(wireframeMAT, 'wire_mat')
+mat.par.colorr = 1.0; mat.par.colorg = 0.0; mat.par.colorb = 0.0
+mat.par.linewidth = 3
+
+# 2. Geometry COMP
+geo = root.create(geometryCOMP, 'my_geo')
+geo.par.rx.expr = 'absTime.seconds * 30'
+geo.par.ry.expr = 'absTime.seconds * 45'
+geo.par.material = mat.path  # NOTE: 'material' not 'mat'
+
+# 3. Shape inside the geo
+box = geo.create(boxSOP, 'cube')
+box.par.sizex = 1.5; box.par.sizey = 1.5; box.par.sizez = 1.5
+
+# 4. Camera
+cam = root.create(cameraCOMP, 'cam1')
+cam.par.tx = 0; cam.par.ty = 0; cam.par.tz = 4; cam.par.fov = 45
+
+# 5. Render TOP
+render = root.create(renderTOP, 'render1')
+render.par.outputresolution = 'custom'
+render.par.resolutionw = 1280; render.par.resolutionh = 720
+render.par.bgcolorr = 0; render.par.bgcolorg = 0; render.par.bgcolorb = 0
+render.par.camera = cam.path
+render.par.geometry = geo.path
+
+# 6. Output null
+out = root.create(nullTOP, 'out1')
+out.inputConnectors[0].connect(render.outputConnectors[0])
+```
+
+**Key rules:**
+- Class names: `wireframeMAT` not `wireframeMat` (all-caps suffix)
+- Geometry SOPs/POPs go INSIDE the geo comp
+- Material: `geo.par.material` not `geo.par.mat`
+- Render geometry: `render.par.geometry = geo.path` (string path)
+- `wireframeMAT.par.wireframemode = 'topology'` for clean wireframe (vs `'tesselated'` for triangle edges)
+- Alternative: Use `renderTOP.par.overridemat` instead of per-geo material
+
+## Feedback TOP
+
+### Basic Structure
+
+```
+input (initial state) ──┐
+                        ├──→ feedback_top ──→ processing ──→ null_out
+                        │                                        ↑
+                        └── par.top = 'null_out' ────────────────┘
+```
+
+### Setup Pattern
+
+```python
+# 1. Processing chain
+glsl = root.create(glslTOP, 'sim')
+null_out = root.create(nullTOP, 'null_out')
+glsl.outputConnectors[0].connect(null_out.inputConnectors[0])
+
+# 2. Feedback referencing null_out
+feedback = root.create(feedbackTOP, 'feedback')
+feedback.par.top = 'null_out'
+
+# 3. Black initial state
+const_init = root.create(constantTOP, 'const_init')
+const_init.par.colorr = 0; const_init.par.colorg = 0; const_init.par.colorb = 0
+
+# 4. Wire: initial → feedback, feedback → processing
+feedback.inputConnectors[0].connect(const_init)
+glsl.inputConnectors[0].connect(feedback)
+
+# 5. Reset to apply initial state
+feedback.par.resetpulse.pulse()
+```
+
+### Common Errors
+
+| Error | Cause | Solution |
+|-------|-------|----------|
+| "Not enough sources specified" | No input connected | Connect initial state TOP |
+| Unexpected initial pattern | Wrong initial state | Use Constant TOP (black) |
+
+### Tips
+
+1. Use float format for simulations: `glsl.par.format = 'rgba32float'`
+2. Reset after setup: `feedback.par.resetpulse.pulse()`
+3. Match resolutions — feedback, processing, and initial state must match
+4. Soft boundary prevents edge artifacts:
+   ```glsl
+   float edge = 3.0 * texel.x;
+   float bx = smoothstep(0.0, edge, uv.x) * smoothstep(0.0, edge, 1.0 - uv.x);
+   float by = smoothstep(0.0, edge, uv.y) * smoothstep(0.0, edge, 1.0 - uv.y);
+   value *= bx * by;
+   ```
+
+### Use Cases
+- **Wave Simulation** — R=height, G=velocity, black initial state
+- **Cellular Automata** — white=alive, black=dead, random noise initial state
+- **Trail / Motion Blur** — blend current frame with feedback, black initial

optional-skills/creative/touchdesigner-mcp/references/pitfalls.md

@@ -143,20 +143,20 @@ Creating nodes with the same names you just destroyed in the SAME script causes
 ```python
 # td_execute_python:
 for c in list(root.children):
-    if c.valid and c.name.startswith('promo_'):
+    if c.valid and c.name.startswith('my_'):
         c.destroy()
-# ... then create promo_audio, promo_shader etc. in same script → CRASHES
+# ... then create my_audio, my_shader etc. in same script → CRASHES
 ```
 
 **CORRECT (two separate calls):**
 ```python
 # Call 1: td_execute_python — clean only
 for c in list(root.children):
-    if c.valid and c.name.startswith('promo_'):
+    if c.valid and c.name.startswith('my_'):
         c.destroy()
 
 # Call 2: td_execute_python — build (separate MCP call)
-audio = root.create(audiofileinCHOP, 'promo_audio')
+audio = root.create(audiofileinCHOP, 'my_audio')
 # ... rest of build
 ```
 
@@ -361,21 +361,13 @@ win.par.winopen.pulse()
 
 `out.sample(x, y)` returns pixels from a single cook snapshot. Compare samples with 2+ second delays, or use screencapture on the display window.
 
-### 32. Audio-reactive GLSL: dual-layer sync pipeline
+### 32. Audio-reactive GLSL: TD-side pipeline
 
-For audio-synced visuals, use BOTH layers for maximum effect:
+For audio-synced visuals: AudioFileIn → AudioSpectrum(timeslice=True, fftsize='256') → Math(gain=5) → choptoTOP(par.chop=math, layout='rowscropped') → GLSL input. The shader samples `sTD2DInputs[1]` at different x positions for bass/mid/hi. Record the TD output with MovieFileOut.
-
-**Layer 1 (TD-side, real-time):** AudioFileIn → AudioSpectrum(timeslice=True, fftsize='256') → Math(gain=5) → choptoTOP(par.chop=math, layout='rowscropped') → GLSL input. The shader samples `sTD2DInputs[1]` at different x positions for bass/mid/hi. Record the TD output with MovieFileOut.
-
-**Layer 2 (Python-side, post-hoc):** scipy FFT on the SAME audio file → per-frame features (rms, bass, mid, hi, beat detection) → drive ASCII brightness, chromatic aberration, beat flashes during the render pass.
-
-Both layers locked to the same audio file = visuals genuinely sync to the beat at two independent stages.
 
 **Key gotcha:** AudioFileIn must be cued (`par.cue=True` → `par.cuepulse.pulse()`) then uncued (`par.cue=False`, `par.play=True`) before recording starts. Otherwise the spectrum is silent for the first few seconds.
 
-### 33. twozero MCP: benchmark and prefer native tools
+### 33. twozero MCP: prefer native tools
-
-Benchmarked April 2026: twozero MCP with 36 native tools. The old curl/REST method (port 9981) had zero native tools.
 
 **Always prefer native MCP tools over td_execute_python:**
 - `td_create_operator` over `root.create()` scripts (handles viewport positioning)
@@ -425,13 +417,16 @@ TD can show `fps:0` in `td_get_perf` while ops still cook and `TOP.save()` still
 
 **a) Project is paused (playbar stopped).** TD's playbar can be toggled with spacebar. The `root` at `/` has no `.playbar` attribute (it's on the perform COMP). The easiest fix is sending a spacebar keypress via `td_input_execute`, though this tool can sometimes error. As a workaround, `TOP.save()` always works regardless of play state — use it to verify rendering is actually happening before spending time debugging FPS.
 
-**b) Audio device CHOP blocking the main thread.** An `audiooutCHOP` with an active audio device can consume 300-400ms/s (2000%+ of frame budget), stalling the cook loop at FPS=0. Fix: keep the CHOP active but set `volume=0` to prevent the audio driver from blocking. Disabling it entirely (`active=False`) may also work but can prevent downstream audio processing CHOPs from cooking.
+**b) Audio device CHOP blocking the main thread (MOST COMMON).** An `audiodeviceoutCHOP` with `active=True` can consume 300-400ms/s (2000%+ of frame budget), stalling the cook loop at FPS=0. **`volume=0` is NOT sufficient** — the audio driver still blocks. Fix: `par.active = False`. This completely stops the CHOP from interacting with the audio driver. If you need audio monitoring, enable it only during short playback checks, then disable before recording.
+
+Verified April 2026: disabling `audiodeviceoutCHOP` (`active=False`) restored FPS from 0 to 60 instantly, recovering from 2348% budget usage to 0.1%.
 
 Diagnostic sequence when FPS=0:
-1. `td_get_perf` — check if any op has extreme CPU/s
+1. `td_get_perf` — check if any op has extreme CPU/s (audiodeviceoutCHOP is the usual suspect)
-2. `TOP.save()` on the output — if it produces a valid image, the pipeline works, just not at real-time rate
+2. If audiodeviceoutCHOP shows >100ms/s: set `par.active = False` immediately
-3. Check for blocking CHOPs (audioout, audiodevin, etc.)
+3. `TOP.save()` on the output — if it produces a valid image, the pipeline works, just not at real-time rate
-4. Toggle play state (spacebar, or check if absTime.seconds is advancing)
+4. Check for other blocking CHOPs (audiodevin, etc.)
+5. Toggle play state (spacebar, or check if absTime.seconds is advancing)
 
 ### 39. Recording while FPS=0 produces empty or near-empty files
 
@@ -484,9 +479,20 @@ If `td_write_dat` fails, fall back to `td_execute_python`:
 op("/project1/shader_code").text = shader_string
 ```
 
-### 42. td_execute_python does NOT return stdout or print() output
+### 42. td_execute_python DOES return print() output — use it for debugging
 
-Despite what earlier versions of pitfall #33 stated, `print()` and `debug()` output from `td_execute_python` scripts does NOT appear in the MCP response. The response is always just `(ok)` + FPS/error summary. To read values back, use dedicated inspection tools (`td_get_operator_info`, `td_read_dat`, `td_read_chop`) instead of trying to print from within a script.
+`print()` statements in `td_execute_python` scripts appear in the MCP response text. This is the correct way to read values back from scripts. The response format is: printed output first, then `[fps X.X/X] [N err/N warn]` on a separate line.
+
+However, the `result` variable (if you set one) does NOT appear verbatim — use `print()` for anything you need to read back:
+```python
+# CORRECT — appears in response:
+print('value:', some_value)
+
+# WRONG — not reliably in response:
+result = some_value
+```
+
+For structured data, use dedicated inspection tools (`td_get_operator_info`, `td_read_chop`) which return clean JSON.
 
 ### 43. td_get_operator_info JSON is appended with `[fps X.X/X]` — breaks json.loads()
 
@@ -496,13 +502,203 @@ clean = response_text.rsplit('[fps', 1)[0]
 data = json.loads(clean)
 ```
 
-### 44. td_get_screenshot is asynchronous — returns `{"status": "pending"}`
+### 44. td_get_screenshot is unreliable — returns `{"status": "pending"}` and may never deliver
 
-Screenshots don't complete instantly. The tool returns `{"status": "pending", "requestId": "..."}` and the actual file appears later. Wait a few seconds before checking for the file. There is no callback or completion notification — poll the filesystem.
+Screenshots don't complete instantly. The tool returns `{"status": "pending", "requestId": "..."}` and the actual file may appear later — or may NEVER appear at all. In testing (April 2026), screenshots stayed "pending" indefinitely with no file written to disk, even though the shader was cooking at 8-30fps.
 
-### 45. Recording duration is manual — no auto-stop at audio end
+**Do NOT rely on `td_get_screenshot` for frame capture.** For reliable frame capture, use MovieFileOut recording + ffmpeg frame extraction:
+```bash
+# Record in TD first, then extract frames:
+ffmpeg -y -i /tmp/td_output.mov -t 25 -vf 'fps=24' /tmp/td_frames/frame_%06d.png
+```
+
+If you need a quick visual check, `td_get_screenshot` is worth trying (it sometimes works), but always have the recording fallback. There is no callback or completion notification — if the file doesn't appear after 5-10 seconds, it's not coming.
+
+### 45. Heavy shaders cook below record FPS — many duplicate frames in output
+
+A raymarched GLSL shader may only cook at 8-15fps even though MovieFileOut records at 60fps. The recording still works (TD writes the last-cooked frame each time), but the resulting file has many duplicate frames. When extracting frames for post-processing, use a lower fps filter to avoid redundant frames:
+```bash
+# Extract at 24fps from a 60fps recording of an 8fps shader:
+ffmpeg -y -i /tmp/td_output.mov -t 25 -vf 'fps=24' /tmp/td_frames/frame_%06d.png
+```
+Check actual cook FPS with `td_get_perf` before committing to a long recording. If FPS < 15, the output will be a slideshow regardless of the recording codec.
+
+### 46. Recording duration is manual — no auto-stop at audio end
 
 MovieFileOut records until `par.record = False` is set. If audio ends before you stop recording, the file keeps growing with repeated frames. Always stop recording promptly after the audio duration. For precision: set a timer on the agent side matching the audio length, then send `par.record = False`. Trim excess with ffmpeg as a safety net:
 ```bash
 ffmpeg -i raw.mov -t 25 -c copy trimmed.mov
+```
+
+### 47. AudioFileIn par.index stays at 0 in sequential mode — not a reliable progress indicator
+
+When `audiofileinCHOP` is in `playmode=2` (sequential), `par.index.eval()` returns 0.0 even while audio IS actively playing and the spectrum IS receiving data. Do NOT use `par.index` to check playback progress in sequential mode.
+
+**How to verify audio is actually playing:**
+- Read the spectrum CHOP values via `td_read_chop` — if values are non-zero and CHANGE between reads 1-2s apart, audio is flowing
+- Read the audio CHOP itself: non-zero waveform samples confirm the file is loaded and playing
+- `par.play.eval()` returning True is necessary but NOT sufficient — it can be True with no audio flowing if cue is stuck
+
+### 48. GLSL shader whiteout — clamp audio spectrum values in the shader
+
+Raw spectrum values multiplied by Math CHOP gain can produce very large numbers (5-20+) that blow out the shader's lighting, producing flat white/grey. The shader MUST clamp audio inputs:
+
+```glsl
+float bass = texture(sTD2DInputs[1], vec2(0.05, 0.25)).r;
+bass = clamp(bass, 0.0, 3.0);   // prevent whiteout
+mids = clamp(mids, 0.0, 3.0);
+hi = clamp(hi, 0.0, 3.0);
+```
+
+Discovered when gain=10 produced ~0.13 (too dark) during quiet passages but gain=50 produced ~9.4 (total whiteout). Fix: keep gain=10, use `highfreqboost=3.0` on AudioSpectrum, clamp in shader.
+
+### 49. Non-Commercial TD records at 1280x1280 (square) — always crop in post
+
+Even with `resolutionw=1280, resolutionh=720` on the GLSL TOP, Non-Commercial TD may output 1280x1280 to MovieFileOut. Always check dimensions with ffprobe and crop during extraction:
+
+```bash
+# Center-crop from 1280x1280 to 1280x720:
+ffmpeg -y -i /tmp/td_output.mov -t 25 -r 24 -vf "crop=1280:720:0:280" /tmp/frames/frame_%06d.png
+```
+
+Large ProRes files (1-2GB) at 1280x1280 decode at ~3fps, so 25s of footage takes ~3 minutes to extract.
+
+## Advanced Patterns (pitfalls 51+)
+
+### 51. Connection syntax: use `outputConnectors`/`inputConnectors`, NOT `outputs`/`inputs`
+
+```python
+# CORRECT
+src.outputConnectors[0].connect(dst.inputConnectors[0])
+# WRONG — raises IndexError or AttributeError
+src.outputs[0].connect(dst.inputs[0])
+```
+
+For feedback TOP, BOTH are required:
+```python
+fb.par.top = target.path
+target.outputConnectors[0].connect(fb.inputConnectors[0])
+```
+
+### 52. moviefileoutTOP `par.input` doesn't resolve via Python in TD 2025.32460
+
+Setting `moviefileoutTOP.par.input` programmatically does NOT work. All forms fail silently with "Not enough sources specified."
+
+**Workaround — frame capture + ffmpeg:**
+```python
+out = op('/project1/out')
+for i in range(300):
+    delay = i * 5
+    run(f"op('/project1/out').save('/tmp/frames/f_{i:04d}.png')", delayFrames=delay)
+# Then: ffmpeg -y -framerate 30 -i /tmp/frames/f_%04d.png -c:v prores -pix_fmt yuv420p /tmp/output.mov
+```
+
+### 53. Batch frame capture — use `me.fetch`/`me.store` for state across calls
+
+```python
+start = me.fetch('cap_frame', 0)
+for i in range(60):
+    frame = start + i
+    op('/project1/out').save(f'/tmp/frames/frame_{str(frame).zfill(4)}.png')
+me.store('cap_frame', start + 60)
+```
+Call 5 times for 300 frames. Each picks up where the last left off.
+
+### 54. GLSL TOP pixel shader requirements in TD 2025
+
+```glsl
+// REQUIRED — declare output
+layout(location = 0) out vec4 fragColor;
+
+void main() {
+    vec3 col = vec3(1.0, 0.0, 0.0);
+    fragColor = TDOutputSwizzle(vec4(col, 1.0));
+}
+```
+**Built-in uniforms available:** `uTDOutputInfo.res` (vec4), `uTDTimeInfo.seconds`, `sTD2DInputs[N]`.
+**Auto-created DATs:** `name_pixel`, `name_vertex`, `name_compute` textDATs with example code.
+
+### 55. TOP.save() doesn't advance time — identical frames in tight loops
+
+`.save()` captures the current cooked frame without advancing TD's timeline:
+```python
+# WRONG — all frames identical
+for i in range(300):
+    op('/project1/out').save(f'frames/f_{i:04d}.png')
+
+# CORRECT — use run() with delayFrames
+for i in range(300):
+    delay = i * 5
+    run(f"op('/project1/out').save('frames/f_{i:04d}.png')", delayFrames=delay)
+```
+**NEVER use `time.sleep()` in TD** — it blocks the main thread and freezes the UI.
+
+### 56. Feedback loop masks input changes — force switch during capture
+
+With feedback TOP opacity 0.7+, the buffer dominates output. Switching input produces nearly identical frames.
+
+**Fix — force switch index per capture:**
+```python
+for i in range(300):
+    idx = (i // 8) % num_inputs
+    delay = i * 5
+    run(f"op('/project1/vswitch').par.index={idx}; op('/project1/out').save('f_{i:04d}.png')", delayFrames=delay)
+```
+
+### 57. Large td_execute_python scripts fail — split into incremental calls
+
+10+ operator creations in one script cause timing issues. Split into 2-4 calls of 2-4 operators each. Within one call, `create()` handles work immediately. Across calls, `op('name')` may return `None` if the previous call hasn't committed.
+
+### 58. MCP instance reconnection after project.load()
+
+`project.load(path)` changes the PID. After loading, call `td_list_instances()` and use the new `target_instance`. For TOX files: import as child comp instead (doesn't disconnect).
+
+### 59. TOX reverse-engineering workflow
+
+```python
+comp = root.loadTox(r'/path/to/file.tox')
+comp.name = '_study_comp'
+for child in comp.children:
+    print(f'{child.name} ({child.OPType})')
+# Use td_get_operators_info, td_read_dat, check custom params
+```
+
+### 60. sliderCOMP naming — TD appends suffix
+
+TD auto-renames: `slider_brightness` → `slider_brightness1`. Always check names after creation.
+
+### 61. create() requires full operator type suffix
+
+```python
+# CORRECT
+proj.create('audiofileinCHOP', 'audio_in')
+proj.create('glslTOP', 'render')
+
+# WRONG — raises "Unknown operator type"
+proj.create('audiofilein', 'audio_in')
+proj.create('glsl', 'render')
+```
+
+### 62. Reparenting COMPs — use copyOPs, not connect()
+
+Moving COMPs with `inputCOMPConnectors[0].connect()` fails. Use copy + destroy:
+```python
+copied = target.copyOPs([source])  # preserves internal wiring
+source.destroy()
+# Re-wire external connections manually after the move
+```
+
+### 63. Slider wiring — expressionCHOP with op() expressions crashes TD
+
+```python
+# CRASHES TD — don't do this
+echop = root.create(expressionCHOP, 'slider_ctrl')
+echop.par.chan0expr = 'op("/project1/controls/slider_brightness1").par.value0'
+
+# WORKING — parameterCHOP as bridge
+pchop = root.create(parameterCHOP, 'slider_vals')
+pchop.par.ops = '/project1/controls'
+pchop.par.parameters = 'value0'
+pchop.par.custom = True
+pchop.par.builtin = False
 ```

optional-skills/creative/touchdesigner-mcp/references/postfx.md

@@ -0,0 +1,183 @@
+# Post-FX Reference
+
+Bloom, CRT scanlines, chromatic aberration, and feedback glow patterns for live visual work.
+
+---
+
+## Bloom
+
+### Built-in Bloom TOP
+
+TD's `bloomTOP` is the fastest path — GPU-accelerated, no shader needed.
+
+```python
+bloom = root.create(bloomTOP, 'bloom1')
+bloom.par.threshold = 0.6     # Luminance threshold (0-1)
+bloom.par.size = 0.03         # Spread radius (0-1)
+bloom.par.strength = 1.5      # Bloom intensity
+bloom.par.blendmode = 'add'   # 'add' or 'screen'
+```
+
+**Audio reactive bloom:**
+```python
+bloom.par.strength.mode = ParMode.EXPRESSION
+bloom.par.strength.expr = "op('audio_env')['envelope'][0] * 3.0 + 0.5"
+```
+
+### GLSL Bloom (More Control)
+
+For multi-pass bloom with color tinting:
+
+```glsl
+// bloom_pixel.glsl — pass1: threshold + tint
+out vec4 fragColor;
+uniform float uThreshold;
+uniform vec3 uBloomColor;
+
+void main() {
+    vec4 col = texture(sTD2DInputs[0], vUV.st);
+    float luma = dot(col.rgb, vec3(0.299, 0.587, 0.114));
+    float bloom = max(0.0, luma - uThreshold);
+    fragColor = TDOutputSwizzle(vec4(col.rgb * bloom * uBloomColor, col.a));
+}
+```
+
+Then blur with `blurTOP` (size ~0.02-0.05), composite back over source with `addTOP` or `compositeTOP` in Add mode.
+
+---
+
+## CRT / Scanlines
+
+Pure GLSL — create a `glslTOP` and paste into its `_pixel` DAT.
+
+```glsl
+// crt_pixel.glsl
+out vec4 fragColor;
+uniform float uTime;
+uniform float uScanlineIntensity;  // 0.0 - 1.0, default 0.4
+uniform float uCurvature;          // 0.0 - 0.15, default 0.05
+uniform float uVignette;           // 0.0 - 1.0, default 0.8
+
+vec2 curveUV(vec2 uv, float amount) {
+    uv = uv * 2.0 - 1.0;
+    vec2 offset = abs(uv.yx) / vec2(6.0, 4.0);
+    uv = uv + uv * offset * offset * amount;
+    return uv * 0.5 + 0.5;
+}
+
+void main() {
+    vec2 res = uTDOutputInfo.res.zw;
+    vec2 uv = vUV.st;
+
+    // CRT barrel distortion
+    uv = curveUV(uv, uCurvature * 10.0);
+
+    // Kill pixels outside curved screen
+    if (uv.x < 0.0 || uv.x > 1.0 || uv.y < 0.0 || uv.y > 1.0) {
+        fragColor = vec4(0.0, 0.0, 0.0, 1.0);
+        return;
+    }
+
+    vec4 col = texture(sTD2DInputs[0], uv);
+
+    // Scanlines
+    float scanline = sin(uv.y * res.y * 3.14159) * 0.5 + 0.5;
+    col.rgb *= mix(1.0, scanline, uScanlineIntensity);
+
+    // Horizontal noise flicker
+    float flicker = TDSimplexNoise(vec2(uv.y * 100.0, uTime * 8.0)) * 0.03;
+    col.rgb += flicker;
+
+    // Vignette
+    vec2 vig = uv * (1.0 - uv.yx);
+    float v = pow(vig.x * vig.y * 15.0, uVignette);
+    col.rgb *= v;
+
+    fragColor = TDOutputSwizzle(col);
+}
+```
+
+---
+
+## Chromatic Aberration
+
+Splits RGB channels and offsets them along screen axes.
+
+```glsl
+out vec4 fragColor;
+uniform float uAmount;   // 0.001 - 0.02, default 0.006
+
+void main() {
+    vec2 uv = vUV.st;
+    vec2 dir = uv - 0.5;
+
+    float r = texture(sTD2DInputs[0], uv + dir * uAmount).r;
+    float g = texture(sTD2DInputs[0], uv).g;
+    float b = texture(sTD2DInputs[0], uv - dir * uAmount).b;
+    float a = texture(sTD2DInputs[0], uv).a;
+
+    fragColor = TDOutputSwizzle(vec4(r, g, b, a));
+}
+```
+
+**Audio-reactive variant** — spike aberration on beats:
+```glsl
+uniform float uBeat;
+void main() {
+    vec2 uv = vUV.st;
+    vec2 dir = uv - 0.5;
+    float amount = uAmount + uBeat * 0.04;
+    float r = texture(sTD2DInputs[0], uv + dir * amount * 1.2).r;
+    float g = texture(sTD2DInputs[0], uv).g;
+    float b = texture(sTD2DInputs[0], uv - dir * amount * 0.8).b;
+    fragColor = TDOutputSwizzle(vec4(r, g, b, 1.0));
+}
+```
+
+---
+
+## Feedback Glow
+
+Warm persistent trails for glow effects.
+
+```glsl
+out vec4 fragColor;
+uniform float uDecay;     // 0.92 - 0.98 for slow trails
+uniform vec3 uGlowColor;  // tint accumulated feedback
+
+void main() {
+    vec2 uv = vUV.st;
+    vec4 prev = texture(sTD2DInputs[0], uv);  // feedback input
+    vec4 curr = texture(sTD2DInputs[1], uv);  // current frame
+
+    vec3 glow = prev.rgb * uDecay * uGlowColor;
+    vec3 result = max(glow, curr.rgb);
+
+    fragColor = TDOutputSwizzle(vec4(result, 1.0));
+}
+```
+
+**Tips:**
+- `uDecay = 0.95` → medium trail
+- `uDecay = 0.98` → long comet tail
+- Set `glslTOP` format to `rgba16float` for smooth gradients
+
+---
+
+## Full Post-FX Stack
+
+Recommended order:
+
+```
+[scene / composite]
+        ↓
+   bloomTOP          ← luminance threshold bloom
+        ↓
+   glslTOP (chrom)   ← chromatic aberration
+        ↓
+   glslTOP (crt)     ← scanlines + barrel distortion + vignette
+        ↓
+   null_out          ← final output
+```
+
+**Performance note:** Each glslTOP is a full GPU pass. For 1920×1080 at 60fps this stack is comfortably real-time. For 4K, consider downsampling bloom input with `resolutionTOP` first.

optional-skills/creative/touchdesigner-mcp/references/troubleshooting.md

@@ -137,7 +137,7 @@ actual = str(n.width) + 'x' + str(n.height)
 
 ### Config location
 
-`$HERMES_HOME/config.yaml` (defaults to `~/.hermes/config.yaml` when `HERMES_HOME` is unset)
+~/.hermes/config.yaml
 
 ### MCP entry format
 

optional-skills/creative/touchdesigner-mcp/scripts/setup.sh

@@ -8,8 +8,7 @@ OK="${GREEN}✔${NC}"; FAIL="${RED}✘${NC}"; WARN="${YELLOW}⚠${NC}"
 
 TWOZERO_URL="https://www.404zero.com/pisang/twozero.tox"
 TOX_PATH="$HOME/Downloads/twozero.tox"
-HERMES_HOME_DIR="${HERMES_HOME:-$HOME/.hermes}"
+HERMES_CFG="$HOME/.hermes/config.yaml"
-HERMES_CFG="${HERMES_HOME_DIR}/config.yaml"
 MCP_PORT=40404
 MCP_ENDPOINT="http://localhost:${MCP_PORT}/mcp"
 
@@ -18,10 +17,7 @@ manual_steps=()
 echo -e "\n${CYAN}═══ twozero MCP for TouchDesigner — Setup ═══${NC}\n"
 
 # ── 1. Check if TouchDesigner is running ──
-# Match on process *name* (not full cmdline) to avoid self-matching shells
+if pgrep -if "TouchDesigner" >/dev/null 2>&1; then
-# that happen to have "TouchDesigner" in their args. macOS and Linux pgrep
-# both support -x for exact name match.
-if pgrep -x TouchDesigner >/dev/null 2>&1 || pgrep -x TouchDesignerFTE >/dev/null 2>&1; then
     echo -e " ${OK} TouchDesigner is running"
     td_running=true
 else
@@ -69,6 +65,9 @@ if 'twozero_td' not in cfg['mcp_servers']:
     }
     with open(cfg_path, 'w') as f:
         yaml.dump(cfg, f, default_flow_style=False, sort_keys=False)
+    print('added')
+else:
+    print('exists')
 " 2>/dev/null && echo -e " ${OK} twozero_td MCP entry added to config" \
               || { echo -e " ${FAIL} Could not update config (is PyYAML installed?)"; \
                    manual_steps+=("Add twozero_td MCP entry to ${HERMES_CFG} manually"); }