legal-skills

Open-Source AI Agent Skills for Legal Professionals — Global Scope, Multilingual

Production-ready AI skills that work across Claude, Claude Code, OpenCode, Cursor, and 30+ other agent platforms

[🇬🇧 English](./README.md) | [🇮🇱 עברית](./README_HE.md)

What Makes These Skills Special?

Cross-Platform → One skill works in Claude, Claude Code, OpenCode, Cursor, GitHub Copilot, OpenHands, Goose, Codex, and every other agentskills.io-compatible tool Multilingual by Design → 8 language packs at launch (English, Hebrew, Arabic, Russian, Spanish, French, German, Portuguese) — lazy-loaded, no wasted tokens Privacy-First → Runs 100% locally. Automatic sensitive-data detection with masking. No telemetry, no external calls. Environment-Aware → Adapts to whatever extraction tools are installed. Gracefully degrades when tools are missing. Practitioner-Built → Designed by lawyers and legal technologists for real case workflows, not toy demos.

New to Agent Skills? See the Agent Skills overview and Anthropic's skill authoring best practices.

Table of Contents

🎯 Available Skills

🚀 Quick Start — pick the easiest route for your tool

All commands below have been tested end-to-end. If any fails, please [open an issue](https://github.com/chen-friedman/legal-skills/issues).

🎯 Option A — Claude Code (two commands)

Inside Claude Code, run:

/plugin marketplace add chen-friedman/legal-skills
/plugin install legal-skills@chen-friedman

That's it. Restart Claude Code if prompted, then in any case folder ask:

"Map this case folder"

Verify it installed:

/plugin list

You should see legal-skills@chen-friedman with status enabled.

🎯 Option B — Claude.ai (web app) — ZIP upload

1. Download the latest skill ZIP: [mapping-legal-cases.zip](https://github.com/chen-friedman/legal-skills/releases/latest/download/mapping-legal-cases.zip)
2. Go to [claude.ai](https://claude.ai) → Settings → Customize → Skills
3. Click Upload custom skill, select the ZIP
4. Enable it, then in any conversation upload your case files and ask: "Map this case folder."

Available on Free, Pro, Max, Team, and Enterprise plans with code execution enabled.

🎯 Option C — OpenCode

git clone https://github.com/chen-friedman/legal-skills.git
# Windows PowerShell:
New-Item -ItemType SymbolicLink -Path "$HOME\.config\opencode\skills\mapping-legal-cases" -Target "$PWD\legal-skills\skills\mapping-legal-cases"
# macOS / Linux:
ln -s "$PWD/legal-skills/skills/mapping-legal-cases" ~/.config/opencode/skills/mapping-legal-cases

🎯 Option D — Cursor / Copilot / Gemini CLI / OpenHands / Goose / Kiro / Roo Code / any other agentskills.io-compatible tool

git clone https://github.com/chen-friedman/legal-skills.git

Then point your tool at legal-skills/skills/mapping-legal-cases/. See your tool's documentation for the exact skill directory path.

🎯 Option E — No-install: paste this prompt into any AI tool

If you can't install skills in your tool, copy-paste the skill directly into your chat:

1. Open skills/mapping-legal-cases/SKILL.md raw on GitHub
2. Paste the full contents into your AI chat (Claude, ChatGPT, Gemini, etc.) with this prompt:

I'll give you a skill definition. Read and internalize it, then apply it to my case folder.
Confirm you've loaded the skill, ask for the case folder path, then run the mapping workflow.

--- SKILL DEFINITION START ---
[paste SKILL.md contents here]
--- SKILL DEFINITION END ---

This won't be as efficient as a native skill install, but it works everywhere.

✅ Verify your environment (after any install method)

python skills/mapping-legal-cases/scripts/extract.py --preflight --pretty

This tells you which extractors are available on your machine (PDF, Word, Excel, OCR, transcription). The skill adapts automatically to whatever's installed.

🔍 mapping-legal-cases

The flagship skill. Point your AI at a folder of case documents — get a structured analysis brief.

What goes in

Any mix of:

• 📄 PDFs (native or scanned)
• 📝 Word documents (.docx, .doc, .odt, .rtf)
• 📊 Spreadsheets (.xlsx, .xls, .csv, .tsv)
• ✉️ Emails (.eml, .msg)
• 📑 Markdown, plain text, JSON, HTML
• 🖼️ Images (auto-OCR if tesseract installed)
• 🔊 Audio (auto-transcribe if whisper installed)
• 🎥 Video (auto-transcribe + metadata)

What comes out

A .casebase/ subfolder with 10 structured documents, every finding cited back to its source:

Plus MAPPING_LOG.md — audit trail listing processed files and what still needs OCR/transcription.

How it works

Two-wave architecture keeps token usage low:

Wave 1 — extract every file once using scripts/extract.py, cache the text, build keyword index and privacy audit.

Wave 2 — four parallel analysis tasks read from the cache to build the 10 outputs. On platforms with subagent support, these run in parallel for speed.

The extractor auto-adapts to whatever's installed:

• Baseline (zero installs): text, markdown, CSV, JSON, .eml emails, zip listings
• With Python packages: PDFs, Word, Excel, images, Outlook emails
• With CLI tools: pandoc (richer formats), tesseract (OCR), ffmpeg (media), whisper (transcription)
• Missing tools? Files are flagged [PENDING OCR] / [PENDING TRANSCRIPTION] — never blocked

Example output

See skills/mapping-legal-cases/references/output-templates.md for the exact structure of each of the 10 generated documents. Real end-to-end demo folder coming in v1.1 with proper PDFs, DOCX and emails.

🌍 Language Packs

Keyword packs are lazy-loaded based on detected document language — no tokens wasted on languages your case doesn't use.

Each pack includes ~150 legal terms across 13 categories (dates, money, parties, procedures, courts, document types, criminal, family, property, labor, insurance, privacy patterns, filename signals) plus jurisdiction-specific privacy regex patterns.

🔒 Privacy & Security

• Runs entirely locally. No document content leaves your machine (unless you install an optional cloud backend).
• Automatic sensitive-data scan. National IDs, credit cards, bank accounts, medical record numbers, passport numbers, phone numbers — all detected and masked in PRIVACY_FLAGS.md.
• Privacy patterns per jurisdiction: US SSN/EIN, Israeli Teudat Zehut, Brazilian CPF/CNPJ, Spanish DNI/NIE, Mexican CURP/RFC, Argentine CUIT, Chilean RUT, German Steuer-ID, French NIR, Russian ИНН/СНИЛС, Portuguese NIF, and more.
• Cache is local-only by default. .casebase/.cache/ contains raw extracted text — treat it like the original files.
• No telemetry, no tracking, no external calls. Audit scripts/extract.py yourself — one self-contained file.

See SECURITY.md for vulnerability reporting.

📦 Optional Dependencies

The skill works with zero dependencies for plain text / CSV / JSON / Markdown / .eml emails. Install what you need for richer formats:

# Python packages (core)
pip install pymupdf pdfplumber python-docx openpyxl Pillow

# Optional — Outlook .msg support
pip install extract-msg

# Optional — audio/video transcription
pip install faster-whisper

External tools (OS-specific):

Run the preflight check any time:

python skills/mapping-legal-cases/scripts/extract.py --preflight --pretty

🗺️ Roadmap

v1.1 — More Languages + Better OCR

• Italian, Dutch, Polish, Turkish, Chinese, Japanese, Korean
• Built-in EasyOCR fallback
• Installation wizard

v1.2 — Structured Extraction Upgrades

• Named-entity recognition (NER) for auto-detecting people, organizations, locations
• Table-aware parsing for financial tables and contract schedules
• Signature detection (signed vs. unsigned documents)
• Version grouping (v1, v2, final)
• Redaction detection

v1.3 — Deeper Jurisdictional Knowledge

• Regional sub-packs (en-us, en-uk, es-mx, fr-ca, de-at, de-ch)
• Jurisdictional procedure packs (FRCP for US federal court, etc.)
• Case-number format library

v1.4 — Workflow Integrations

• Companion skill: drafting-legal-responses
• Companion skill: legal-research
• Calendar integration (DEADLINES.md → iCal / Google Calendar)
• Redaction companion (auto-redact based on PRIVACY_FLAGS.md)

v2.0 — Advanced Analysis

• Cross-case search
• Precedent detection
• Conflict-of-interest check
• Settlement valuation heuristics
• Multi-lingual corpus handling

Full details → ROADMAP.md

🤝 Contributing

Contributions are welcome — especially new language packs!

High-value contribution areas:

1. Language packs — Italian, Dutch, Polish, Turkish, Chinese, Japanese, Korean, Hindi, and more
2. Regional sub-packs — es-mx, en-uk, fr-ca, etc.
3. New file type handlers — .numbers, .pages, .key
4. Real-world feedback — from practicing lawyers across jurisdictions

See CONTRIBUTING.md for the full guide.

📚 Related Projects

Check out awesome-legaltech — a curated list of open-source Legal AI & LegalTech tools, datasets, and communities.

Relevant building blocks used by this skill:

• PyMuPDF — PDF extraction
• Pandoc — Universal document converter
• Tesseract — OCR engine
• faster-whisper — Speech-to-text
• Agent Skills standard — Cross-platform skill format

📄 License

Apache License 2.0. See LICENSE and NOTICE.

Commercial use allowed. Fork it, modify it, ship it in your product — just keep the attribution notice.

👤 Author

Chen Friedman — founder, Lawcal AI LegalTech / GovTech AI consultancy, Israel 🇮🇱

• GitHub: @chen-friedman
• Also check out: awesome-legaltech — curated open-source legal tech

🌟 Support

If this helps you:

• ⭐ Star the repo
• 🐛 File issues for bugs or feature requests
• 🌐 Contribute a language pack for your jurisdiction
• 📢 Share with colleagues who could use it

Questions? Open a discussion or see SECURITY.md for vulnerability reporting.

Built with ❤️ for legal professionals worldwide.