English hub of method-neutral legal AI skills for Claude Code / Cowork. The verification core, content-quality tools and EU-level law sources port across jurisdictions; deep national law stays where it belongs.
Maintainer: Wieslaw Mazur / MateMatic Solutions. Curatorial licence: MIT (skills keep their own licences declared in SKILL.md).
The method/substance line. What ports to English is the method - grounding, red-team review, scoring, citation discipline, EU law search. What does not port is the substance of a specific legal system. Deep Polish-jurisdiction skills (case law, KRS, redlining, PL grounding) live in the Polish hub: awesome-matematic-skills-pl.
- Domain bundles installed with one command - 18 skills grouped by function into five plugins (plus the nine-connector multi-jurisdiction bundle). The verification foundation installs without any connector; EU law sources ship their MCP connector with the bundle.
- Curated list - links to strong open skills from other teams (below). We point to them, we do not republish them.
- Companion Polish hub for Polish-jurisdiction skills and live PL sources.
Two ways to install. A (npx skills) works in any agent that supports the Agent Skills format (Cursor, OpenAI Codex, Windsurf, Gemini CLI, Claude Code) and installs individual skills. B is the native Claude Code marketplace, which installs whole bundles and keeps each plugin's inline CLAUDE.md.
What to install. Start with verification-foundation (the jurisdiction-neutral core) and add content-quality, data-protection, ai-governance and eu-law-sources as needed. Installing the whole hub with npx skills add ... (no --skill) pulls every skill at once; for most users it is cleaner to cherry-pick with --skill or install one bundle at a time with /plugin.
# Whole hub - every skill across the bundles, discovered recursively by SKILL.md
npx skills add matematicsolutions/awesome-matematic-skills-en
# Only the skills you want - cherry-pick by frontmatter name
npx skills add matematicsolutions/awesome-matematic-skills-en --skill citation-extraction-en humanizer-ennpx skills (vercel-labs/skills) fetches SKILL.md files and installs them into the agent's config directory (.claude/skills/ or .agents/skills/). It runs anywhere the Agent Skills format runs - not only Claude Code. This route installs the skills only. The MCP connector in eu-law-sources needs the plugin config, so install it with method B.
# 1. Add the marketplace (once)
/plugin marketplace add matematicsolutions/awesome-matematic-skills-en
# 2. Install the bundles you need
/plugin install verification-foundation@matematic-skills-en # core, no connectors
/plugin install content-quality@matematic-skills-en # English-writing tools
/plugin install eu-law-sources@matematic-skills-en # EU law + eu-sparql connector
/plugin install ai-governance@matematic-skills-en # AI Act / NIS2 / privilege pre-flightThe verification foundation and content-quality bundles use no connectors and send nothing out. eu-law-sources runs its MCP connector via npx, so it needs node in the environment.
Install-always. No connectors, sends no data out. /plugin install verification-foundation@matematic-skills-en.
| Skill | What it does | Licence |
|---|---|---|
| adversarial-legal-review-en | Red team for a high-stakes deliverable - builder/attacker/synthesizer/verifier finds the weakness before opposing counsel does. | Apache-2.0 |
| legal-syllogism-en | Structure legal reasoning as major premise / minor premise / conclusion; surface hidden premises. | Apache-2.0 |
| citation-extraction-en | Extract and structure legal citations from text so they can be verified, not trusted on sight. | Apache-2.0 |
| clause-checklist-en | Check a contract against a clause checklist; flag missing or risky clauses. | Apache-2.0 |
| output-scoring-en | Score an AI legal output against a rubric before it ships. | Apache-2.0 |
| opposing-counsel-attack-en | Single-pass opposing-counsel attack: core theory of attack, steel-manned-then-X-rayed reconstruction, point-by-point vulnerabilities. The cheap tier below the full adversarial review. | Apache-2.0 |
| judicial-first-impression-en | How a submission lands on a sceptical judge reading it cold under time pressure - a seven-part neutral assessment, no rewriting, no attacking. | Apache-2.0 |
/plugin install content-quality@matematic-skills-en.
| Skill | What it does | Licence |
|---|---|---|
| humanizer-en | Strip AI-writing patterns (inflated symbolism, rule of three, em-dash overuse, AI vocabulary); make text read naturally. | MIT |
| reviewer-en | Grumpy senior editor: a verdict (disaster/weak/mediocre/ok) and charges anchored to file:line. Points out what is wrong, does not rewrite. |
MIT |
Ships the eu-sparql MCP connector (read-only, public EUR-Lex / Cellar). Requires node/npx. /plugin install eu-law-sources@matematic-skills-en.
| Skill | What it does | Licence |
|---|---|---|
| eu-sparql-search | Search EU legislation and CJEU case law via the Publications Office SPARQL endpoint (CELEX, ELI URI, multi-language acts). | Apache-2.0 |
| authority-triage-en | Route the source layer before answering: EU primary/secondary law, national chain, courts, soft law - which layer governs, in what order to verify, through which connector. | Apache-2.0 |
Operational GDPR tooling grounded in the regulation's articles and EDPB guidelines, no connectors. Each ends in a draft for decision; the outward act (filing, sending, erasure, signing) stays human. /plugin install data-protection@matematic-skills-en.
| Skill | What it does | Licence |
|---|---|---|
| gdpr-dpia-en | Data Protection Impact Assessment: threshold test (EDPB 9 criteria), Art. 35(7) structure, Art. 36 prior consultation. Deterministic threshold screener. | Apache-2.0 |
| gdpr-breach-72h-en | 72h breach response: decision tree, risk assessment, SA notification (Art. 33) with a deadline counter, communication to data subjects (Art. 34). Deterministic 72h deadline calculator. | Apache-2.0 |
| gdpr-dsar-en | Data subject rights (DSAR): classify Art. 15-22 rights, Art. 12(3) deadline counter, exemption/refusal gates, draft response + register. Deterministic one-month deadline calculator (Reg. 1182/71). | Apache-2.0 |
| gdpr-ropa-dpa-en | Records of processing (RoPA, Art. 30) + processor-contract review (DPA, Art. 28(3)(a)-(h)) with a redline of missing clauses. Deterministic Art. 28 clause check. | Apache-2.0 |
Regulatory triage for organisations deploying or building AI. Drafts for decision; the outward act (registering, notifying, filing, sending) stays human. /plugin install ai-governance@matematic-skills-en.
| Skill | What it does | Licence |
|---|---|---|
| eu-ai-act-triage-en | EU AI Act triage (Reg. 2024/1689): AI-system definition, Art. 5 prohibitions, high-risk (Annex I/III + Art. 6(3) filter), GPAI, Art. 50 transparency, role mapping (provider/deployer/importer/distributor, Art. 25), FRIA signal (Art. 27). Ends in a classification card. | Apache-2.0 |
| nis2-compliance-triage-en | NIS2 triage (Dir. 2022/2555): essential/important entity scope with the size-cap rule, the ten Art. 21(2) measures, Art. 23 incident clock (24h/72h/1 month), Art. 20 management duties, national transposition check. | Apache-2.0 |
| privilege-preflight-en | Pre-flight before sending legal content to an external AI surface: SAFE/CAUTION/STOP band on privilege, professional secrecy, GDPR and trade-secret factors, plus a redacted-safe draft for human approval. | Apache-2.0 |
We point to these, we do not republish them - check each repo's own licence before use.
| Skill | Source |
|---|---|
| Legal skills hub, 139 skills (licences per-skill in frontmatter) | lawvable/awesome-legal-skills |
| Swiss legal source & authority triage (MIT) | lawvable/awesome-legal-skills |
| French judicial reasoning, ENM methodology (MIT) | lawvable/awesome-legal-skills |
| AI Act for engineers | morellid/ai-act-skill |
| Anthropic legal suite | anthropics/claude-for-legal |
| AI Act resources | GenAI-Gurus/awesome-eu-ai-act |
| Presidio - PII anonymisation | microsoft/presidio |
| GRC compliance | Sushegaad/Claude-Skills-Governance-Risk-and-Compliance |
| Privacy procedures | mukul975/Privacy-Data-Protection-Skills |
| Contract review (CUAD) | evolsb/claude-legal-skill |
EU law ships with eu-law-sources above. For national law, see the eu-legal-mcp connector line - one MCP connector per jurisdiction (DE/AT/ES/FI/IE/NL/SE/FR/LU), each grounding acts and case law against the official source.
- You install only what applies. Grounding, red-team review and scoring travel to any jurisdiction; country-specific case law does not, so this hub never ships you Polish or German rules you cannot use.
- GDPR-safe by default. The verification and content skills run locally and send no data out. Connectors are read-only over public APIs.
- AI Act Art. 12 + Art. 50. Record-keeping and the transparency duty are built into how these skills present output - every output is a draft, marked as AI-generated.
- Vendor-agnostic. The skills follow the open Agent Skills format and run wherever it is implemented.
All skills inherit three shared standards in references/: citation style and confidence tags, responsibility and data protection, firm deployment. Each plugin's CLAUDE.md carries the core inline so it works after installing the plugin alone.
- Curatorial (README, taxonomy, marketplace.json): MIT
- Per-skill: declared in
SKILL.mdfrontmatter (Apache-2.0 for MateMatic-authored, MIT for adaptations) - Curated entries above: each keeps its own upstream licence
matematicsolutions.com | LinkedIn Wieslaw Mazur | github.com/matematicsolutions