epd-parser
Extract GWP, life-cycle stages, certifications, and impact metrics from an EPD PDF. Use when given a declaration to parse; not to find or compare EPDs.
AlpacaLabsLLC · AI review on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add AlpacaLabsLLC/skills-for-architects --skill epd-parserAbout this skill
name: epd-parser description: Extract GWP, life-cycle stages, certifications, and impact metrics from an EPD PDF. Use when given a declaration to parse; not to find or compare EPDs. allowed-tools:
- Read
- Write
- Edit
- Bash
- Glob
- Grep
- AskUserQuestion
/as:epd-parser — EPD PDF Parser
Harness note: use
/as:<skill>on Claude Code and$<skill>on Codex. Resolve<skill-root>as the directory containing this loadedSKILL.mdand<plugin-root>as the plugin root that containsskills/, and use equivalent native tools when host tool names differ.
Extract structured environmental impact data from EPD (Environmental Product Declaration) PDF files. Uses PyMuPDF for text extraction and Claude's reasoning to parse varying EPD formats into a standardized 42-column schema.
EPDs follow ISO 14025 / ISO 21930 / EN 15804 and report life cycle environmental impacts of building products. This skill reads those PDFs and structures the data for comparison, specification, and LEED documentation.
Input
The user provides EPD PDFs in one of these ways:
- File paths — one or more PDF file paths
- Folder path — a directory containing PDFs (will process all
.pdffiles) - Just invoked — ask the user for file paths or a folder
Output Schema
Use the canonical 42-column contract in schema/epd-schema.md. It is separate from the 33-column FF&E product schema. Never substitute or persist an FF&E-shaped record as EPD data. Leave unavailable values empty and use a plain URL for EPD Link.
Material Category Vocabulary
Use ONE normalized term: Concrete, Steel, Aluminum, Wood/Timber, Insulation, Gypsum, Glass, Ceramic/Tile, Carpet, Resilient Flooring, Roofing Membrane, Sealant, Paint/Coating, Masonry, Stone, Composite Panel, Acoustic, Cladding, Rebar, Cement, Aggregate, Furniture, Other.
EPD-specific data in Notes (col AO)
EPDs contain fields that don't have dedicated columns. Append these to Notes:
- EPD Type:
Type: Product-specificorType: Industry-averageorType: Sector - Verification:
Verified by: [verifier name] - EN version:
EN 15804+A1orEN 15804+A2(important for comparability) - Source File:
Source: holcim-readymix-epd.pdf - LCA Software:
LCA: GaBiorLCA: SimaProorLCA: openLCA - Biogenic carbon note: If the EPD reports biogenic carbon storage separately
Example Notes cell: Type: Product-specific | Verified by: Underwriters Laboratories | EN 15804+A2 | LCA: GaBi | Source: holcim-readymix-epd.pdf
LEED Eligibility Logic (col AI)
- Yes: Product-specific, third-party verified EPD conforming to ISO 14025
- Partial: Industry-average or sector EPD (counts for LEED Option 1 but not Option 2)
- No: Self-declared, unverified, or non-conforming
Workflow
Step 1: Get input
Parse the user's input to identify PDF file(s) and output preferences.
- If given a folder, list all
.pdffiles and report count - If no PDFs found or path is invalid, ask the user
- Report: "Found N EPD PDF(s) to process."
Step 2: Extract text from PDF
Use PyMuPDF (fitz) to extract text from each PDF. Run this Python script via Bash:
import fitz
import sys
import json
pdf_path = sys.argv[1]
doc = fitz.open(pdf_path)
pages = []
for i, page in enumerate(doc):
text = page.get_text()
pages.append({"page": i + 1, "text": text})
doc.close()
print(json.dumps({"filename": pdf_path.split("/")[-1], "total_pages": len(pages), "pages": pages}))
For each PDF, extract all pages and keep the JSON output transiently for parsing; do not create a persistent JSON artifact.
Step 3: Parse EPD data
Read the extracted text and identify all environmental impact data. This is the core intelligence step.
For small EPDs (<=30 pages): Process all pages at once.
For large EPDs (>30 pages): Process in chunks of 15 pages. Carry forward context between chunks.
Parsing instructions:
-
Identify the EPD structure — EPDs typically have these sections:
- General information (pages 1-2): manufacturer, product, declared unit, program operator, validity
- Product description (pages 2-3): materials, manufacturing process, application
- LCA information: system boundary, data sources, allocation rules
- Impact indicator tables: the core data — usually 1-3 pages of tables
- Resource use and waste tables
- Additional information: scenarios, biogenic carbon, recycled content
-
Extract product identity first — manufacturer, product name, declared unit, functional unit. These are always on page 1-2.
-
Extract EPD metadata — registration number, program operator, PCR, standard, dates, system boundary. Usually on page 1 or in a header/sidebar.
-
Find and parse impact indicator tables — This is the most critical step:
- Look for tables with life cycle stage columns (A1, A2, A3 or A1-A3 combined, A4, A5, B1-B7, C1-C4, D)
- Impact categories are rows: GWP, ODP, AP, EP, POCP, ADP (or ADPE/ADPF)
- EN 15804+A2 EPDs split GWP into: GWP-total, GWP-fossil, GWP-biogenic, GWP-luluc. Capture all that exist.
- EN 15804+A1 EPDs report a single GWP row. Put this value in GWP-total (A1-A3).
- If A1, A2, A3 are reported separately (not combined), sum them for the A1-A3 total.
- Units are standardized: GWP = kg CO2e, ODP = kg CFC-11e, AP = kg SO2e (or mol H+ eq for +A2), EP = kg PO4e (or mol N eq / kg P eq for +A2), POCP = kg C2H4e (or kg NMVOC eq for +A2).
-
Extract resource use — PERE, PENRE, fresh water, waste. Usually in a separate table following the impact indicators.
-
Extract additional data — recycled content %, manufacturing plant, country of origin, LCA software, verifier name.
-
Determine LEED eligibility — based on EPD type and verification status.
-
Leave fields blank rather than guessing — if a field isn't in the EPD, leave