Chat Order OCR for Openclaw

1. Ensure Python is available and locate the bundled CLI:

CLI=/home/ubuntu/skills/chat-order-ocr/scripts/order_ocr_cli.py
python "$CLI" --help

2. For screenshot OCR, install Tesseract and the Traditional Chinese and English language data using your operating system's package manager. Confirm the language packs:

tesseract --list-langs

The output should include chi_tra and eng. Install the corresponding Tesseract language packages if either is missing.

3. Prepare a UTF-8 text file, OCR text stream, or supported image files. Run a basic text-to-JSON conversion:

python "$CLI" --input messages.txt --format json --output orders.json

4. Parse OCR text from stdin and export a grouped CSV:

cat ocr.txt | python "$CLI" --source ocr --group-by person --format csv --output orders.csv

5. Process multiple screenshots with local Tesseract OCR and an optional product catalog:

python "$CLI" --images screenshots/*.jpg --catalog catalog.json --format json --output ocr-orders.json

6. Use an image directory for supported formats or print sorted JSON to standard output:

python "$CLI" --images-dir screenshots/ --format json --output batch-orders.json
python "$CLI" --input messages.txt --sort name-desc --format json

7. Before delivery or fulfillment, manually inspect every row marked confidence: check. OCR output is not automatically approved, and the skill must not fabricate quantities or fill in missing order details.

JSON envelope

Order row metadata

• person: The identified ordering person when available.
• item: The original product text captured by the parser.
• quantity: The detected numeric or Chinese quantity.
• source: The input origin, such as text or ocr.
• confidence: Usually high for clear text-source rows and check for OCR, ambiguous, incomplete, or otherwise uncertain results.
• matched_item: The standardized catalog item when alias matching succeeds.
• match_status: 已對應 for a successful catalog match or 未對應，請確認 when no match is found.

Catalog format

Provide a JSON object mapping standard product names to aliases:

{
  "Standard Product": ["Alias One", "Alias Two"]
}

Catalog matching removes whitespace, ignores letter case, recognizes aliases contained in the input, and prefers longer aliases. It is a deterministic alias lookup rather than semantic inference, so the original item text remains available for review.

CSV output

CSV exports use UTF-8 BOM encoding for spreadsheet compatibility and include customer name, item, quantity, source, and confidence fields. Rows requiring manual review must remain visibly marked rather than being silently corrected.