contact-vcard-extractor

name

contact-vcard-extractor

description

Extract contact information from text, chat logs, web pages, screenshots, business card photos, and QR codes/OCR results; clean names, phone numbers, email addresses, companies, job titles, addresses, and websites; generate vCard/.vcf files; and guide users to import them into iOS Contacts or share them with others. Trigger words include: extract contacts, save to contacts, generate vCard/.vcf, business card recognition, add contacts from images/screenshots/text, share contacts, and export contacts.

Contact vCard Extractor

Objective

Convert contact information from user-provided text or images into previewable, importable, and shareable .vcf vCard files. Prioritize the real user experience: extract the data and have the user confirm it before generating the file. Mark any uncertain fields and do not silently fill in guesses.

Typical Workflow

1. Receive Input
   • Text: Directly parse user messages, pasted text, or results extracted from web pages.
   • Images: Business card photos, screenshots, posters, or chat screenshots. First use apple-vision ocr to recognize text. If it appears to contain a QR code, use apple-vision barcode.
   • Files: First check <attachments>/, <workspace>/, and <mounts>/.
2. Extract Fields
   • Name FN/N
   • Phone TEL: Multiple numbers must be split into separate TEL entries. The number field should contain only the number itself. Do not append labels such as "front desk," "mobile," or "WeChat" to the number. Export all phone numbers uniformly as TEL;TYPE=CELL; users can change the type themselves if needed. Labels can be shown in the summary and, if necessary, placed in notes.
   • Email EMAIL
   • Company ORG
   • Job title TITLE
   • Address ADR
   • Website URL
   • Notes NOTE: Source, WeChat ID, uncategorized but long-term useful information, and uncertain recognition items. Do not include temporary to-dos, reminders, or next follow-up items.
3. User Confirmation
   • List the fields concisely and highlight content that may have been recognized incorrectly.
   • If the name or phone/email is missing, ask whether the user wants to add it. If the user is in a hurry, generate an "Unnamed Contact."
4. Generate vCard
   • Use the bundled script: ~/.agents/skills/contact-vcard-extractor/scripts/contact_to_vcard.py
   • Output to <workspace>/Contact Name.vcf. Remove special characters from the file name. If necessary, use contact.vcf.
5. Present and Import/Share
   • Provide a Markdown file link: [Import Contact](<workspace>/xxx.vcf).
   • Use open <workspace>/xxx.vcf to preview/share within the app.
   • If the user explicitly wants to open the import screen, run apple-open <workspace>/xxx.vcf or open. Usually, prefer open to stay in the chat.

Image/OCR Command Pattern

apple-vision ocr <attachments>/card.jpg --lang zh-Hans,en --level accurate --compact
apple-vision barcode <attachments>/card.jpg --compact

After saving the OCR output as a text file, run the parsing script.

Text-to-vCard Command

python3 ~/.agents/skills/contact-vcard-extractor/scripts/contact_to_vcard.py \
  --text-file <workspace>/contact_ocr.txt \
  --out <workspace>/contact.vcf \
  --json

Text can also be passed via stdin. Do not inline very long text in shell commands; for long text, first use file_write to write it to a file.

User Experience Details

• Do not include temporary information in contact notes: For example, to-dos such as "Send a quote next Tuesday," "Call back tomorrow," or "Follow up at the end of the month" should be removed from the vCard notes. In the reply, separately ask, "Would you like me to create a reminder or to-do item?" If the user explicitly agrees, use apple-reminders create to create the reminder.
• Phone fields must be clean and consistently CELL: TEL may contain only numbers, such as 010-66668888 or 13344445555. For "010-66668888 (front desk), mobile [phone]," split it into two phone entries. Use "front desk/mobile" only as display labels or notes, and do not append them to the number. When exporting, use TEL;TYPE=CELL for all phone numbers. Do not write types such as VOICE/HOME/WORK unless the user explicitly specifies them.
• Do not import directly into Contacts unless the user explicitly confirms. First generate the vcf and have the user open it to confirm.
• Keep privacy prompts lightweight: Contacts are personal information. Remind users to confirm authorization and content only when sharing or processing in batches.
• Multiple contacts: If the text or image clearly contains multiple people, generate separate .vcf files, or merge them into one contacts.vcf file containing multiple vCards. Finally, list each person in a table.
• QR codes: If the QR code content is MECARD:, BEGIN:VCARD, tel:, mailto:, WeChat, or a URL, parse it according to the content. Raw vCard content can be saved directly as .vcf; MECARD must be converted.
• Chinese names: The vCard N field can use the first character as the family name and the remaining characters as the given name. If uncertain, prioritize correct display in FN.
• International numbers: Preserve +Country Code, extensions, and spaces. Do not forcefully rewrite them.
• File naming: Prefer Name.vcf; if the name is empty, use contact-YYYYMMDD-HHMM.vcf.
• Final reply format:
  1. A one-sentence confirmation that it has been generated.
  2. A summary of the fields.
  3. A link to the file.

Sample Reply

I’ve organized this business card:

Field	Content
Name	Zhang San
Phone	[phone]
Email	[e-mail]
Company	Example Tech

[Import/Share Contact](/张三.vcf)

After opening it, you can add it to Contacts or share it directly with someone else.

Bundled Script Notes

contact_to_vcard.py performs basic rule-based extraction and vCard escaping. It is not the only method: for complex input, low-quality OCR, or messy layouts, use model judgment to manually correct the fields before generating the vCard.