How viprasol-tech/invoice-extractor works
This repository packages a portable invoice-extraction agent skill, not a conventional application: an agent uses skill.md to turn a supplied document into validated JSON and a human-readable report. The README describes classification, extraction, normalization, arithmetic reconciliation, confidence and discrepancy reporting. The workflow below reflects those documented stages; the skill source itself was not excerpted, so relationships are evidenced by the README rather than implementation calls.
Main entry points: skill.md Extract fields
Components 11
Document intake
- Invoice skill
skill.md
Extraction and checks
- Classify document
skill.md - Extract fields
skill.md - Normalize values
skill.md - Reconcile totals
skill.md
Review and output
- Assess confidence
skill.md - Flag issues
skill.md - JSON and summary
skill.md
Other components
- User
- Agent platform
- Human reviewer
Connections 11
- User → Agent platform: provides document
README.md - Agent platform → Invoice skill: uses skill
README.md - Invoice skill → Classify document: guides classification
README.md - Classify document → Extract fields: classifies for extraction
README.md - Extract fields → Normalize values: normalizes fields
README.md - Normalize values → Reconcile totals: checks arithmetic
README.md - Reconcile totals → Flag issues: flags failures
README.md - Extract fields → Assess confidence: surfaces uncertainty
README.md - Flag issues → JSON and summary: reports issues
README.md - Assess confidence → JSON and summary: includes confidence
README.md - JSON and summary → Human reviewer: returns for review
README.md