Finance OCR Pro

Run this skill only after OCR intent from the user.

This skill is especially helpful for financial reports, annual reports, prospectuses, investor presentations, regulatory filings, research reports, and other documents with complicated structure, charts, graphs, tables, and mixed layout elements.

Security And Privacy

Before running OCR, make the operating model clear:

• - This skill requires three environment variables, all of which must be configured before OCR can run:

• - OCR sends rendered page images and structured prompts to BASE_URL. This is the primary data-transmission path. Users must verify that the endpoint is trusted before processing sensitive documents.
• If the user wants offline or local-only OCR, BASE_URL must point to a local VLM service. Do not run this skill against an external endpoint with sensitive documents unless the provider is trusted.
• Never commit a populated .env file. Use .env.example as a template and keep real credentials local.

Pre-Run Notice

After the user asks for OCR or extraction, give a short notice that includes:

• - whether BASE_URL is local or remote
• which VLM_MODEL will be used
• which execution mode will be used
• where results will be written
• that page images and prompts will be transmitted to the configured endpoint

Proceed automatically unless the user asks to change those defaults.

Defaults To Announce

• - Running mode: background job by default
• Model: INLINECODE9
• Threads: INLINECODE10
• Result path:

Setup

Use the skill-local virtual environment if present.

• - macOS/Linux: INLINECODE13
• Windows: INLINECODE14
• Fallback: INLINECODE15

Run:

CODEBLOCK0

If setup is incomplete, run:

CODEBLOCK1

Preferred Execution

By default, start a background worker:

CODEBLOCK2

Then inspect progress and outputs:

CODEBLOCK3

Use synchronous mode only when the user explicitly wants inline execution:

CODEBLOCK4

Notes

• - Inputs: PDF, common office documents, Apple office formats, and images.
• Outputs: merged Markdown, HTML review report, DOCX, and Excel.
• OCR requires API_KEY, BASE_URL, and VLM_MODEL to be configured before running.
• Sensitive document pages are transmitted to the configured endpoint during OCR unless the endpoint is a local service.
• Best suited for financial documents and other visually dense materials with tables, charts, graphs, and complex page structure.
• Office-document conversion may require LibreOffice.
• OCR extraction by the VLM model may be time-consuming; check the status regularly.