Receipt Compliance Assistant

Problem

In corporate reimbursement workflows, invoice recognition, authenticity checks, template entry, and approval submission often remain manual and fragmented. Hand-entry causes field errors, low-quality photos drop key fields, and DingTalk, WeCom, or Feishu approvals can fail because of missing API permissions or incorrect flow codes. This skill moves those steps into a local Python workflow, with a focus on keeping invoice data from leaving the machine.

How It Works

• OCR recognition: reads PNG, JPG, TIFF, and PDF files, then emits structured JSON; PDF processing requires poppler.
• Authenticity checks: generates a State Taxation Administration verification link from invoice code and number, with optional third-party verification engines.
• Reimbursement and approvals: fills company Excel templates and submits approvals to DingTalk, WeCom, or Feishu after API keys are configured.
• Accounting output: maps recognized fields to accounting subjects, prepares voucher summaries, and generates bookkeeping voucher files.
• Risk and reconciliation: flags sequential invoice numbers, round amounts, frequent issuance, and abnormal product names, then matches bank statement CSV rows against invoice data.

Boundaries

The skill does not provide tax advice and does not guarantee the accuracy of third-party verification results. OCR quality depends on resolution, glare, and invoice flatness, while handwritten and fixed-amount invoices are not fully supported. Approval submissions are generally not reversible, so platform permissions, flow codes, and API keys must be verified first. Batch runs record failed files, which makes manual review a sensible step for high-stakes reimbursements.