What problem does it solve?
This Skill enables open-ended extraction from multimodal evidence (images, PDFs, documents) to identify and structure all relevant information without relying on fixed slots, accelerating evidence analysis.
Core Features & Use Cases
- Open-ended extraction: Identify and extract all meaningful information from evidence content, regardless of format.
- EvidenceCard output: Produce results in the EvidenceCard data structure, supporting both joint (grouped) and individual cards.
- Use Case: When reviewing a batch of image receipts and chat screenshots, quickly surface key facts like parties, amounts, dates, and items.
Quick Start
Provide one or more evidence materials as URLs with an evidence_id, for example:
evidence_materials:
- url: "https://example.com/evidence1.jpg"
evidence_id: "uuid-1"
- url: "https://example.com/evidence2.jpg"
evidence_id: "uuid-2"
Output:
EvidenceCard data structure:
joint card example: card_type: "微信聊天记录", card_is_associated: true, card_features: [
{ "slot_name": "Example Group", "slot_value_type": "group", "slot_value": null, "confidence": 1.0, "reasoning": "Grouped by identifier", "image_sequence_info": [{"evidence_id": "uuid-1", "sequence_number": 1}, {"evidence_id": "uuid-2", "sequence_number": 2}], "sub_features": [{"slot_name": "Example Slot", "slot_value": "Value", "slot_value_type": "string", "confidence": 0.95, "reasoning": "Derived from content", "reference_evidence_ids": ["uuid-1", "uuid-2"]}]}
]
single card example: card_type: "身份证", card_is_associated: false, card_features: [
{ "slot_name": "姓名", "slot_value": "John Doe", "slot_value_type": "string", "confidence": 0.98, "reasoning": "Detected on document", "slot_group_info": null },
{ "slot_name": "公民身份号码", "slot_value": "123456789012345678", "slot_value_type": "string", "confidence": 0.98, "reasoning": "Detected on document", "slot_group_info": null }
]
