通过图片识别 receipt、订单或发票内容，提取商家、日期、金额和商品明细， 并转换为可入账的结构化记录。

This skill appears coherent for turning receipt images into ledger rows, but consider these practical points before installing: - Privacy: receipt images can contain personal and financial data. Confirm how the platform's image model (yunwu/gpt-5-mini) processes and stores images (do they leave your environment, how long are they retained?). - Accuracy: OCR and classification can be imperfect—review parsed records (amounts, dates, categories) before posting to your accounts. The skill defaults a missing date to the current system date; verify that behavior matches your bookkeeping rules. - Confirm expectations for ambiguous items: the skill asks to prompt the user when uncertain; ensure your agent is configured to do that rather than auto-accepting uncertain data. - Minimal system use: it runs a date command when the image lacks a date—this is harmless but note it invokes a simple system command. - Testing: run the skill on representative sample receipts to confirm category mappings, currency formatting, and multi-item handling meet your needs. If any of the above is a concern (especially data residency of image processing), seek clarification from the skill author or avoid sending sensitive receipts until you confirm handling policies.

Type: OpenClaw Skill Name: ledger-image-entry Version: 1.0.0 The skill is designed to extract transaction details from receipt images and format them into a structured ledger. It uses a standard shell command (`date`) in SKILL.md to retrieve the current date if it's missing from the image, which is a legitimate functional requirement. No malicious patterns, data exfiltration, or prompt injection attacks were identified.