| name | ifc-extract |
| description | Extract data from IFC building models. Accepts element types (door, wall, window, slab, roof, etc.) and optional property filters. Translates natural language queries into CLI commands. Results saved to output/ for cross-validation. |
| allowed-tools | Bash, Read, Write, Grep |
| argument-hint | [file.ifc] [element-type] [--property NAME] [--value VALUE] |
/ifc-extract — IFC Data Extraction
Extract building element data from IFC models. For full CLI command syntax, flags, type shorthands, and natural language translation tables, see /cli-tools.
Step 1: Find the IFC file
If no file path provided:
ls data/*.ifc 2>/dev/null
Multiple files → ask user. No files → tell user to place IFC file in data/.
Step 2: Run the extraction
Scripts are at skills/cli-tools/scripts/ifc_extract.py. Always include --save.
python skills/cli-tools/scripts/ifc_extract.py <file.ifc> <command> [args] --save
Use /cli-tools to look up the correct command for the user's request. Common patterns:
| User says | Command |
|---|
| "show me all doors" | list door --save |
| "fire-rated doors" | query door --property Pset_DoorCommon.FireRating --not-null --save |
| "external walls" | query wall --property Pset_WallCommon.IsExternal --value true --save |
| "how many windows per level" | quantities window --group-by storey --save |
| "what's in this model" | summary --save |
| "data quality check" | validate --save |
Step 3: Present results
- Summarize key findings in plain language ("Found 47 doors, 12 fire-rated, 35 without")
- Show a formatted table for first 10-20 elements
- Highlight notable items: missing properties, null fire ratings, proxy elements
- Never compute totals yourself — use
quantities command
- Confirm save location
Step 4: Offer follow-ups
- After
list door: "Filter for fire-rated doors? Check quantities per level?"
- After
summary: "Which element type to explore?"
- After any extraction: "Run
/pdf-extract to find the matching PDF schedule?"
Rules
- LLMs never do arithmetic. Use
quantities for counts, areas, volumes.
- Null properties are normal. Export settings issue, not data error.
- Always use
--save for downstream use by /cross-validate and /report.
