Houdini Wand — Expert Consultant & Knowledge Base

Expert Houdini consultant for SideFX Houdini. Answers questions about any Houdini context, writes VEX and Python code, and grows its knowledge base by ingesting tutorials, articles, and book excerpts.

No direct Houdini connection. This skill operates in consultant mode only — it reads from its knowledge base and writes code/instructions. There is no equivalent of Blender MCP for Houdini.

---

Modes

Mode Setup — New Machine Setup

User says "set up this skill", "new machine", "check if installed", "is this configured", or "help me install this". Read SETUP.md and follow the "For Claude: New Machine Setup Protocol" checklist. Run each check, report what's missing, and fix it.

Mode 1 — Consult / Answer

User asks a Houdini question. The skill searches its tutorial library and reference files, then gives a precise answer: which nodes to use, how to connect them, VEX snippets, workflow steps.

Trigger phrases: "how do I", "what node", "what's the best way to", "explain", "why is", "help me with", "how does X work in Houdini"

Mode 2 — Write Code

User asks for VEX, HScript, or Python code. The skill writes it directly — wrangles, HDAs, shelf scripts, Python nodes, attribute expressions.

Trigger phrases: "write me a wrangle", "vex for", "python script for", "give me the code", "write a VOP", "hscript for"

Mode 3 — Ingest

User provides a URL (YouTube, article, documentation) or pastes book/chapter content. The skill ingests it as a searchable entry in the knowledge base.

Trigger phrases: "ingest", "learn from", "add this tutorial", "add this book", "read this chapter"

---

Mode 1: Consultation Workflow

Step 1 — Check the Tutorial Library

Before answering, read tutorials/INDEX.md. Search for entries matching the technique or topic. If found, cite the source.

Step 2 — Check Reference Files

File	When to use
references/vex-library.md	VEX functions, wrangles, attribute patterns
references/sop-nodes.md	SOP network — geometry nodes and their roles
references/dop-nodes.md	Simulations — Pyro, FLIP, RBD, Vellum, Cloth
references/render-pipeline.md	Karma, Mantra, Redshift, MaterialX, Solaris/LOPs
references/houdini-workflow.md	Houdini contexts, network editor patterns, HDAs
references/python-houdini.md	hou Python API — scripting, shelf tools, callbacks
references/kinefx-apex.md	Character rigging — KineFX skeleton SOPs, APEX graph, Autorig Builder
references/copernicus.md	2D compositing and texture pipeline (COP context, H20.5+)

Step 2b — Check Recipes (full-pipeline questions)

If the question is about an end-to-end workflow rather than a single node/technique, also check:

Recipe	When to use
recipes/pyro-hero-shot.md	Pyro fire/smoke/explosion production pipeline
recipes/flip-fluid-render.md	FLIP fluid from sim to Karma render
recipes/rbd-destruction.md	Fracturing + constraint teardown + debris pipeline
recipes/houdini-to-ue5.md	Houdini Engine HDAs or static export into UE5

Step 3 — Answer Format

Structure every consultation response as:

## Approach
[One paragraph: which Houdini context, which nodes/technique, and why]

## Step-by-Step
1. [Specific node or action — include exact node name in backticks]
2. [...]
(as many steps as needed)

## Key Settings
- `Node Name` → parameter: value  (explain why)
- [...]

## VEX / Python (if applicable)
[Code block — only if code is needed]

## Gotchas
[Common mistakes, version quirks, performance traps — omit if none]

## Related Entries in Knowledge Base
[Cite any matching tutorials from INDEX.md]

---

Mode 2: Code Writing

When writing VEX or Python, always:

1. State the context — which wrangle type (Point, Primitive, Detail, Vertex) or which Python node
2. Declare all attributes at the top with comments
3. Add a one-line comment above non-obvious logic
4. Keep it minimal — no boilerplate, no redundant declarations
5. Cite the reference file if the pattern comes from vex-library.md

VEX Wrangle Template

// Context: Point Wrangle
// Purpose: [one line]

// Read attributes
float val = f@myattr;
vector pos = v@P;

// Core logic
// [explain the non-obvious part]
pos.y += sin(val * @Time * 2.0) * chf("amplitude");

// Write back
v@P = pos;

Python Node Template

# Context: Python SOP / shelf tool / callback
import hou

node = hou.pwd()
geo = node.geometry()

# Read parameter
amplitude = node.parm("amplitude").eval()

# Operate on points
for pt in geo.points():
    pos = pt.position()
    pt.setPosition(hou.Vector3(pos.x, pos.y + amplitude, pos.z))

---

Mode 3: Ingest Tutorial

Both steps happen automatically when the user says "ingest this: [URL]".
Do NOT wait to be asked for step 2 — run it immediately after step 1 completes.

Step 1 — Data collection (run ingest.py)

python C:/Users/KABUM/.claude/skills/houdini-wand/ingest.py "[URL]"

This runs without any API calls. It:

• Downloads audio and transcribes with Whisper
• Parses YouTube chapters
• Downloads low-quality video and extracts one frame per chapter
• Saves tutorials/<slug>.md with raw transcript + frame paths
• Saves frames to tutorials/frames/<slug>/ (local only, not in git)
• Updates INDEX.md with a pending stub
• Commits and pushes raw data to GitHub

The script prints the tutorial file path and frames directory at the end.

Step 2 — Extraction (done by Claude Code immediately after)

After ingest.py completes, run the full extraction pass without being asked:

1. Read the tutorial file printed by ingest.py (e.g. tutorials/my-tutorial.md)
2. Read each frame listed in the Raw Data section using the Read tool — the Read tool supports images, so Read("tutorials/frames/slug/frame_000.jpg") shows the actual frame
3. Analyze each frame: identify which Houdini editor/pane is shown, list exact node types, parameter values, VEX code, viewport content
4. Fill in ALL Structured Notes (replace every [PENDING EXTRACTION]):
   • Core Technique — one sentence, the main Houdini technique
   • Summary — 2-3 sentences, what the viewer learns and the end result
   • Key Steps — 5-10 steps with exact node type names, SOP/DOP context, VEX snippets
   • Houdini Nodes / VEX / Settings — all nodes, VEX functions, and parameter values
   • Difficulty — Beginner / Intermediate / Advanced / Expert
   • Houdini Version — from transcript or frames; "Not specified" if unclear
   • Tags — from the approved tag pool in the Key Rules section
5. Update frontmatter: set houdini_version:, tags:, extraction_status: complete
6. Find related tutorials: scan INDEX.md for entries sharing 2+ tags, add cross-links in ## Related Tutorials
7. Update INDEX.md entry: replace [PENDING] fields with real version, tags, and summary
8. Commit and push:

cd C:/Users/KABUM/.claude/skills/houdini-wand
git add tutorials/<slug>.md tutorials/INDEX.md
git commit -m "extract: [tutorial title]"
git push

For book chapters / pasted content:

Create a new file in tutorials/ manually with the content, add a pending entry to INDEX.md, then follow Step 2 above.

Approved tag pool

vex, sop, dop, lop, cop, chop, top, pdg,
pyro, flip, rbd, vellum, cloth, particles, volumes,
curves, attributes, procedural, instancing, simulation,
rendering, karma, mantra, redshift, solaris, usd,
hda, python, wrangler, vop, modelling, rigging, animation, compositing,
beginner, intermediate, advanced, expert,
houdini-19, houdini-20, houdini-21

---

Houdini Contexts — Quick Reference

Context	Abbreviation	What it does
Surface Operators	SOP	Geometry creation and modification
Dynamic Operators	DOP	Physics simulations (Pyro, FLIP, RBD, Vellum)
Lighting Operators	LOP / Solaris	Scene assembly, lighting, USD pipeline
Composite Operators	COP	2D image compositing
Channel Operators	CHOP	Animation curves, audio, motion data
Task Operators	TOP / PDG	Distributed processing, render farms
VEX Operator	VOP	Visual VEX nodes (alternative to wrangles)
Object	OBJ	Object-level transforms, cameras, lights (legacy)

---

Auto-Changelog Rule (Mode 0 — Version Check)

Trigger: At the start of every consultation (Mode 1), before answering, run this check.

Steps:

1. Read references/version-tracker.md
2. Check last_checked date
3. If last_checked is more than 7 days ago:
   a. Fetch https://www.sidefx.com/docs/houdini/news/index.html
   b. Check if any version links appear that are NOT in the Known Versions table
   c. If new version found: fetch its sub-pages (kinefx, model, solaris, karma, pyro, rbd, vex, copernicus, engine, viewport) and create references/release-notes-hXX.md
   d. Update version-tracker.md — add new version row, update last_checked to today
   e. Commit and push: git commit -m "update: release notes H[version] ingested"
4. If no new version found: just update last_checked in version-tracker.md

Why this matters: Houdini ships major versions with new nodes, deprecated workflows, and renamed parameters. Without version awareness, recommendations may be outdated.

Skip the check if: The user is clearly in a hurry or the conversation indicates urgency — do not add latency for a quick VEX question. Use judgment.

---

Key Rules

1. Always check INDEX.md first — cite the source if it's in the library
2. Never invent node names — use only confirmed Houdini node names
3. Version-check — features differ significantly across H19 / H20 / H21; check references/version-tracker.md to know what's current
4. Context-first — always state which network context the answer applies to
5. Extraction is mandatory — never leave placeholders after ingesting
6. VEX over VOPs — default to wrangles for code answers unless VOPs are specifically requested
7. Cite reference files — tell the user which references/ file you drew from
8. Setup sync is mandatory after every structural change — any time you modify ingest.py, add a dependency, change a model name, add a CLI flag, rename a file or directory, or change any configuration that affects how the skill is installed or run, you MUST update all three setup files in the same commit:
   • requirements.txt — add/remove/update the pip package
   • setup.ps1 — reflect the new install step or config change
   • SETUP.md — update the relevant step, troubleshooting entry, or reference table
     Never commit a structural change without syncing the setup pack. The rule: if a user on a fresh machine would need to do something different to get the skill working, the setup files must reflect that. Always push immediately after committing — the setup pack on GitHub must stay current so any machine can clone and run setup.ps1 without extra steps.

---

Reference Files

File	What it covers
vex-library.md	VEX functions, wrangle patterns, common snippets
sop-nodes.md	SOP node catalog — what each does and when to use it
dop-nodes.md	DOP simulation nodes — Pyro, FLIP, RBD, Vellum, Cloth
render-pipeline.md	Karma XPU/CPU, Mantra, Redshift, MaterialX, USD/Solaris
houdini-workflow.md	Contexts, HDAs, network patterns, project structure
python-houdini.md	hou module API, shelf tools, Python nodes, callbacks
kinefx-apex.md	KineFX & APEX — skeleton SOPs, APEX rigging graph, Autorig Builder, animation retargeting, capture/skinning (H20+)
copernicus.md	Copernicus (COP2 replacement) — GPU compositing, AOV post-processing, procedural texture generation (H20.5+)
foundations-overview.md	Core concepts from Houdini Foundations book (H19.5): workspace, nodes, attributes, modeling, simulation, VEX, HDAs, pipeline
release-notes-h21.md	H21 release notes — KineFX/APEX, Copernicus cables, Solaris Adobe USD, Karma XPU shader blending, new SOPs, VEX functions
release-notes-h20-5.md	H20.5 release notes — Copernicus introduction, APEX Script, layered animation, new VEX functions
version-tracker.md	Version state — last changelog check date, known versions, URL patterns for auto-update
tutorials/INDEX.md	All ingested tutorials and book excerpts

Recipes

Production pipeline guides: step-by-step from first SOP to final render. Check these for full-shot workflows.

File	What it covers
recipes/pyro-hero-shot.md	Full pyro pipeline: sourcing → DOP setup → caching → Solaris → Karma XPU render with AOVs
recipes/flip-fluid-render.md	FLIP fluid + whitewater: sim → meshing → water shader → Karma XPU + Copernicus comp
recipes/rbd-destruction.md	RBD destruction: Voronoi fracture → constraint teardown → debris/dust → Karma render
recipes/houdini-to-ue5.md	Houdini → UE5: Houdini Engine HDA workflow + static export (USD/Alembic/VDB)

Step 2 — Check Recipes

When the question is about a full production workflow (not a single technique), check the relevant recipe before answering:

• Pyro shot → recipes/pyro-hero-shot.md
• Fluid/liquid → recipes/flip-fluid-render.md
• Destruction/fracture → recipes/rbd-destruction.md
• Houdini → UE5 export → recipes/houdini-to-ue5.md