td-trend-scout · git:20260727.48beb34 · 2026-07-27 · sha256 51a9b6e372a2cfc1
td-trend-scout git:20260727.48beb34A
Immutable. This exact content is served forever at /api/v1/blob/51a9b6e372a2cfc1.
--- name: td-trend-scout description: "Scout ONE external surface of the TouchDesigner community for what's HYPED in 2025-2026 that could become a tdmcp tool. Use when a td-trend-scout sub-agent is assigned a surface (community-showcase, tutorials, generative-ai, hardware-interactive, or vfx-aesthetics) during the tdmcp-hype-scout harness — produces `_workspace/hype-scout/01_scout_<surface>.md` with cited trend candidates ranked by hype intensity, recency, and build-ease in tdmcp." --- # td-trend-scout — single-surface hype scouting procedure This skill is loaded by a `td-trend-scout` sub-agent. You receive **one surface** from the orchestrator and produce one file: `_workspace/hype-scout/01_scout_<surface>.md`. You do not synthesize across surfaces — that's `td-hype-synthesizer`'s job. ## Procedure ### Step 1 — Open your scratchpad (immediately) Write the file header before any web work, so a network failure mid-run leaves usable partial output: ```bash mkdir -p _workspace/hype-scout ``` Then `Write` the file with this header (replace `<surface>`): ```markdown # Hype scout — <surface> **Surface:** <surface> **Scout window:** 2025-Q3 → 2026 (preferentially), with documented revivals of older trends allowed. **Run:** <YYYY-MM-DD> ## Trend candidates ``` Append each candidate as you confirm it (don't batch). If your run is cut off, the orchestrator can resume from this file. ### Step 2 — Map your surface Use the surface map below to know **where** to look. Don't search everywhere — go deep on the channels that matter for your surface. #### `community-showcase` — what artists are shipping - TD Forum showcase: `https://forum.derivative.ca/c/community-showcase/` - Instagram: search `#touchdesigner`, `#touchdesignerart`, `#touchdesignerartist`, `#touchdesigner099` - Vimeo: `TouchDesigner` tag, `derivative` group - Behance: `TouchDesigner` projects, filter by Recent - x.com / Twitter: `from:elekktronaut OR from:dotsimulate OR from:paketa12` etc., `#touchdesigner` - Are.na: TD channels (`touchdesigner`, `td-sketches`) Look for: recurring techniques across multiple posts in the last ~6 months; *what they're making*, not who made it. #### `tutorials` — what is being taught - YouTube channels (search their last 6-12 months of uploads): - elekktronaut (Bileam Tschepe) — high-end tutorials, often front-runs trends - Paketa12 — practical effects - The Interactive & Immersive HQ — pro patterns, installations - dotsimulate — AI bridges (StreamDiffusion, ComfyUI) - Crystal Jow — generative + audioreactive - Matthew Ragan — fundamentals + recent advanced work - Acid Boi, Yeyou Studio, Noones Lab - Skillshare / Domestika TD course catalog — what's selling - Recent blog posts: `derivative.ca/wiki`, Medium TD tag Look for: techniques covered by 3+ channels in 6 months = a hot front-running trend. #### `generative-ai` — TD ↔ AI crossover - dotsimulate's TouchDiffusion / StreamDiffusion TOX (search recent demos) - ComfyUI-in-TD bridges (Spout/NDI/OSC, "comfyui touchdesigner") - Realtime ML in TD: depth (MiDaS, Depth Anything), pose (MediaPipe, BlazePose), style transfer, segmentation - ControlNet rigs driven from TD - Audio-to-video AI (Stable Audio + TD, etc.) - LCM / Lightning models real-time - Forum threads on AI integration (`forum.derivative.ca` — Components & Tools, Beginners) #### `hardware-interactive` — physical inputs - LiDAR: Livox HAP/Tele, Ouster OS-1, iPhone LiDAR via Record3D / WebSocket - Depth cameras: RealSense D4xx/L515, Kinect Azure (and post-Azure alternatives) - Hand tracking: Leap Motion 2 / Ultraleap, MediaPipe Hands via NDI/Spout - Body pose: MediaPipe BlazePose, OpenPose, BlazePose-via-NDI - Audio interfaces / Ableton Link / MIDI: trending mappings, TouchOSC profiles, Stream Deck workflows - Sensors: capacitive (Bare Conductive), IR, ToF, eye tracking (Tobii) - Recent installation case studies in showcases #### `vfx-aesthetics` — visual languages For each candidate aesthetic, look across showcase + Instagram for recurrence: - ferrofluid / metaballs / liquid metal / chrome - raymarched volumetrics, SDF scenes - particle systems with feedback loops, slit-scan - datamosh / glitch revival, pixel-sort - kinetic typography, type-as-form - scan-line / CRT / VHS aesthetic - retro-cyber, Y2K revival - point-cloud / 3D-scan / photogrammetry aesthetic - generative gradients, brutalist-data dashboards - low-bit / dithered / Bayer - voronoi / cellular / organic growth - liquid-metal / chrome - isometric pixel / voxel ### Step 3 — For each candidate trend, do the vet pass Before writing an entry, check three things: 1. **Recency & evidence.** Do I have ≥2 citations from 2025-2026 (or documented revival posts)? URLs that resolve, not just memory. If not, drop or downgrade. 2. **Tool shape in tdmcp.** Can I name the layer + the TD operators or bridge endpoint? Use grep against the repo to confirm operators exist in the KB: ```bash grep -irn "realsense\|kinect\|pointCloud" src/knowledge/data/ | head grep -irn "create_.*audio\|create_.*particle" src/tools/ | head grep -irn "<keyword>" src/tools/ src/knowledge/data/ docs/ROADMAP.md ``` 3. **tdmcp coverage.** Search `src/tools/` and `recipes/` for existing equivalents: - NOT-COVERED: nothing close — clean new tool. - PARTIAL: related tool exists but doesn't fully cover the trend (e.g. `create_audio_reactive` covers audio→visual, but not "Ableton-Link locked phrase morph"). - COVERED: it already exists — note the existing tool name; consider as `EXTENSION` if the trend adds a preset/parameter. ### Step 4 — Write the entry Append to your scratchpad in this exact format: ```markdown ### <N>. <Short trend name> - **What artists are doing:** 1-3 sentences, concrete. - **Why it's hyped:** 1 sentence about *why now*. - **Evidence:** - <URL> — <1-line description of what's there> - <URL> — <1-line> - (optional) <URL> — <1-line> - **Hype intensity:** High / Medium / Low - **Recency tag:** e.g. `Q4-2025`, `2026`, `2024-revival` - **Suggested tdmcp tool:** `create_<name>` — Layer <1|2|3> — touches operators `<a, b, c>` (cite KB path or note `UNVERIFIED — probe live`). - **Build-ease guess:** S / M / L (S=hours, M=day, L=multi-day or new bridge endpoint) - **tdmcp coverage:** NOT-COVERED / PARTIAL — existing `<tool>` / COVERED — existing `<tool>` - **Cross-surface footnote:** (optional) e.g. "also appears in `tutorials`" ``` ### Step 5 — Tally and close End the file with a one-line tally: ```markdown --- **Tally:** N candidates — Hype H/M/L: `<x/y/z>` — Coverage NOT/PART/COV: `<x/y/z>` — Build S/M/L: `<x/y/z>` ``` If the file is tagged `PARTIAL-DUE-TO-NETWORK`, say so at the top. ## Hype lenses (use multiple, not just one) When you scan a surface, run each candidate through at least 2 of these lenses to confirm it's really trending: | Lens | Question | |---|---| | **Recurrence** | Is this technique appearing in 3+ recent posts/videos by different creators? | | **Front-runner** | Is a respected tutorial channel (elekktronaut, dotsimulate) teaching it now? | | **Cross-tool** | Is it bridging TD with another hot tool (ComfyUI, Ableton, Notch, Unreal)? | | **Hardware unlock** | Is it riding a new hardware capability (LiDAR cheap now, iPhone depth, M-series GPU)? | | **Aesthetic moment** | Is this visual language showing up across genres (clubs, brand work, art installations)? | | **AI moment** | Is this trend riding the real-time ML wave (StreamDiffusion, Depth Anything, etc.)? | Single-lens trends are usually weak signal — note them but score Hype: Low. Multi-lens trends (e.g. front-runner + cross-tool + AI moment) are strong signal — score Hype: High. ## Citation rules (non-negotiable) - **Real URLs only.** Do not paraphrase from memory. - **At least 2 citations per entry.** If you can only find 1, lower confidence and mark `EVIDENCE-WEAK`. - **Mix sources.** Two posts from the same Instagram account are weaker than one Instagram + one YouTube. - **Date when possible.** Quote upload/post dates in the 1-line description (`"Sept 2025 tutorial"`). ## Working tools - `WebSearch` to find recent posts/videos/threads on your surface. - `WebFetch` to verify a URL is live and pull a 1-line summary. - `Bash` (grep/find) + `Read` to check tdmcp coverage (`src/tools/`, `src/knowledge/data/`, `docs/ROADMAP.md`, `recipes/`). - `Write` (incremental append via `Edit` after the initial header `Write`). If `WebSearch`/`WebFetch` aren't available, do what you can with cached knowledge but tag the file `PARTIAL-DUE-TO-NETWORK` at the top and lower every entry's confidence by one notch. ## What not to do - Don't scout outside your assigned surface (write cross-surface footnotes instead). - Don't pad to a count — 6 high-confidence entries beat 15 weak ones. - Don't claim hype without citations. "Everyone is doing X" without 2 URLs is not an entry. - Don't invent operators. Cite the KB (`src/knowledge/data/`) or mark `UNVERIFIED — probe live`. - Don't synthesize. You scout one surface; the synthesizer dedupes across them. ## Re-invocation If `_workspace/hype-scout/01_scout_<surface>.md` already exists: 1. Read it first. 2. Apply only the requested change — refresh citations, deepen a category, re-check coverage against an updated tree. 3. Preserve existing numbering where it doesn't conflict.