houdini-hda-automation

name	houdini-hda-automation
description	Pipeline skill — HDA library automation and PDG/ROP execution: scan loaded HDA libraries, inspect definitions (inputs, parm templates, sections, version), instantiate digital assets with inputs/parameters, validate cooked nodes, cook TOP/PDG networks, and execute ROP output-driver chains. Pair with houdini-hda for install/save and houdini-render for single-ROP captures.
license	MIT
compatibility	dcc-mcp-houdini 0.1+, Houdini 20.5+, dcc-mcp-core 0.17+
allowed-tools	Bash Read Write Edit
metadata	{"dcc-mcp":{"dcc":"houdini","layer":"domain","stage":"pipeline","version":"1.0.0","tags":["houdini","hda","otl","digital-asset","pdg","top","rop","render","automation","validation","pipeline"],"search-hint":"hda otl digital asset definition instantiate validate cook pdg top tasks rop output driver render chain dependencies automation","tools":"tools.yaml"}}

houdini-hda-automation

Typed automation tools for Houdini Digital Assets (HDA/OTL) and procedural batch execution via PDG/TOP networks and ROP output-driver chains. Complements houdini-hda (which installs/lists/executes/saves HDAs) with library scanning, definition introspection, instantiation, validation, and graph-level cooking.

Tool groups

• hda-query (read-only): scan_hda_libraries, inspect_hda_definition, validate_hda.
• hda-edit: instantiate_hda (async — creates and cooks a node).
• pdg-rop (async): cook_top_network, execute_rop_chain.

Context limitations

• Headless / no license: all tools degrade through structured skill_error envelopes when hou is unavailable; PDG/ROP cooking still requires a valid Houdini session.
• Long runs: instantiate_hda, validate_hda, cook_top_network, and execute_rop_chain are execution: async with timeout_hint_secs set — cooking and rendering can be slow; agents should poll rather than block.
• PDG stats: cook_top_network reports work_item_count on a best-effort basis via getPDGGraphContext(); it may be null when the context is unavailable.
• ROP dependencies: execute_rop_chain honours upstream ROP inputs by default; pass ignore_inputs=true to render only the named driver.

Tracer-bullet flow

1. scan_hda_libraries() → find an installed library / node type
2. inspect_hda_definition("labs::my_asset") → discover inputs + parm names
3. instantiate_hda("/obj", "labs::my_asset", parameters={...}, inputs=[...])
4. validate_hda("/obj/my_asset1") → confirm a clean cook
5. cook_top_network("/obj/topnet1/output") or execute_rop_chain("/out/mantra1", frame_range=[1, 24])