Summary
Adds an entry under Document Processing for the Polish Legal AI Skills hub - a curated 22-skill collection for Polish-jurisdiction legal AI, alongside the existing Master Claude for Legal entry.
What the hub covers
- LLM output validation pipeline - request router, intake sufficiency check, citation grounding (anti-hallucination), adversarial red-team review, fidelity check, AI Act art. 12 audit bundle
- Polish PII anonymization (
let-it-be) - offline, deterministic, with checksum-valid detection for PESEL / NIP / REGON / KRS / IBAN / dowód osobisty, plus heuristic for Polish names, companies, addresses. No data leaves the machine.
- Case law connectors - SAOS REST API for Polish common courts / Supreme Court / Constitutional Tribunal / National Appeal Chamber; EUR-Lex SPARQL for EU acts and CJEU
- Document operations - native Word Track Changes for
.docx (no formatting loss), markdown conversion, PDF reading-order extraction
- Compliance scaffolding - Google Workspace encrypted backup mapped to RODO (GDPR) art. 32; AI Act art. 12 record-keeping bundles with SHA256 manifest
Why this fits the list
- Real production use: validated on real Polish legal proceedings (case-law search, document anonymization before LLM submission, audit bundles)
- Adopts the google/skills architecture patterns (Safety Tiers R/M/D for destructive operations, lean SKILL.md with
references/ offload)
- Vendor-agnostic: works on Claude Code, Claude.ai, Claude Apps SDK
- Apache-2.0 / MIT per skill, curatorial license MIT
Contributor checklist
🤖 Generated with Claude Code