houdini-pipeline

name	houdini-pipeline
description	Pipeline skill — generic, adapter-owned project and shot/package automation: set/query the project root, tag and read portable asset metadata, validate a scene for missing files / bad output dirs / dirty state / cook errors, collect file dependencies into a manifest (no copying), and export a shot/package manifest (frame range, cameras, output nodes, caches, written files). No dependency on private production services.
license	MIT
compatibility	dcc-mcp-houdini 0.1+, Houdini 20.5+, dcc-mcp-core 0.19.45+
allowed-tools	Bash Read Write Edit
metadata	{"dcc-mcp":{"dcc":"houdini","layer":"domain","stage":"pipeline","version":"1.0.0","tags":["houdini","pipeline","project","job","metadata","validation","dependencies","shot","package","manifest","export"],"search-hint":"project job workspace root asset metadata tag validate scene missing files dependencies manifest collect shot package export frame range cameras caches outputs","tools":"tools.yaml"}}

houdini-pipeline

Generic, public pipeline helpers that stay adapter-owned and filesystem-oriented. They deliberately avoid any dependency on private production services — external integrations are left as optional extension points (consume the returned manifests in your own publish step).

Tool groups

• project: set_project, get_project — manage $JOB with explicit paths and structured validation.
• metadata: tag_asset_metadata, get_asset_metadata — store/read a portable JSON payload under the dcc_mcp_meta: user-data key on a node or the hip file.
• validation (read-only): validate_scene, collect_dependencies — surface missing files, bad output dirs, dirty state, cook errors, and a dependency manifest (never copies files).
• package: export_shot_package — manifest of frame range, fps, cameras, output nodes, caches, and written files; optional JSON write.

Context limitations

• No private services: metadata uses a generic, adapter-owned schema; the shot manifest is plain JSON you can feed into any downstream pipeline.
• No implicit copying: collect_dependencies and export_shot_package return manifests only; export_shot_package writes a file solely when write_manifest=true and an output_path is given.
• Heuristic file-parm detection: file references are detected via string parm templates with a FileReference string type, with name-based heuristics (output / sopoutput / picture / lopoutput) to classify outputs.
• Headless friendly: all tools degrade through structured skill_error envelopes when hou is unavailable.

Tracer-bullet flow

1. set_project("/projects/showA/shot010", create=true)
2. tag_asset_metadata({"shot": "sh010", "artist": "me"}, node_path="/obj")
3. validate_scene() → fix any missing_files / bad_output_dirs
4. collect_dependencies() → review the dependency manifest
5. export_shot_package(output_path="/projects/showA/shot010/package.json", write_manifest=true)