Boutique · Skills
The same skills we use at MateMatic. Each is a ready-made procedure: we describe the problem it solves and how. Load once and keep - locally, in PATRON or Claude. Free to start; the catalogue grows steadily. These are language-portable and work on your English text.
Free to start · Local · GDPR-safe · No Git, no technical knowledge
Skills are grouped into domain bundles - install whole sets with one command, in any agent or natively in Claude Code. No Git, no unzipping.
any agent - Cursor, Codex, Claude Code
npx skills add matematicsolutions/awesome-matematic-skills-en
Claude Code - by bundle
/plugin marketplace add matematicsolutions/awesome-matematic-skills-enBundles to choose
/plugin install verification-foundation@matematic-skills-en /plugin install content-quality@matematic-skills-en /plugin install data-protection@matematic-skills-en /plugin install eu-law-sources@matematic-skills-en
Deep Polish-jurisdiction skills stay in the Polish hub. Repository: awesome-matematic-skills-en ↗
A skill is a methodology instruction the agent loads and applies - that's this catalogue. An MCP is a connector that plugs the agent into a data source, live case law or a register (see Connectors). A plugin is an install bundle grouping several skills into one command (the bundles above). Three different things solving three different problems - pick the one that matches what you're missing.
The same skills we use at MateMatic. Each is a ready-made procedure - we describe the problem it solves and how. Free to start; the catalogue grows steadily.
ProblemYou've written a long, source-heavy English article that reads smoothly - and you still can't tell if it holds up.
What it doesA grumpy senior editor delivers a verdict and a list of charges, pointing to the spot in the text. It never suggests fixes and never praises for the sake of it.
ProblemText from AI smells of AI: word inflation, em dashes, the rule of three, empty pathos.
What it doesIt removes the AI tells and leaves natural, precise language - built on Wikipedia's "Signs of AI writing". Keeps your meaning, cuts the filler.
ProblemYou send a high-stakes opinion without knowing where the other side will strike.
What it doesA red team builds the strongest version of your thesis, then attacks it - counter-arguments, contrary authority, procedural gaps - and tells you what survives. Cost-gated, for high-stakes deliverables only.
ProblemEU acts and CJEU rulings are scattered across EUR-Lex and the Cellar repository - finding the exact one by hand is slow.
What it doesQueries the EU Publications Office SPARQL endpoint by CELEX number, date or subject - EU legislation and CJEU rulings come back with an ELI identifier and a source link.
ProblemA flawed opinion rarely fails at the conclusion - it fails at the premise nobody stated.
What it doesLays out the explicit syllogism - rule and its reading, material facts, application of facts to each element, conclusion - and flags the weak links to attack. Maps to civil-law subsumption and common-law IRAC.
Problem"Sounds right" is not "the citations check out" - and you risk sending one for the other.
What it doesScores a finished output in two layers: objective (citations, provisions, dockets - checkable) and subjective (1-5 rubric: correctness, completeness, clarity, jurisdiction, grounding). Returns a decision: send, revise, or full verification.
ProblemYou read what the contract says - but the most expensive clause is often the one that is not there.
What it doesWalks one contract against 41 clause categories (the CUAD taxonomy) and marks each present, absent or risky, quoting the clause when present. Extractive, common-law native.
ProblemA citation you missed is a citation nobody checked.
What it doesMechanically pulls every legal reference (ECLI, CELEX, Official Journal, case names, provisions), normalises them and resolves short references (ibid., supra) to their antecedent. The list feeds verification and EU retrieval.
ProblemYou're deploying a system with personal data and don't know if it needs a DPIA - skipping one is a fine risk.
What it doesRuns the EDPB 9-criteria threshold test, the Art. 35(7) structure, and the Art. 36 prior-consultation decision. Grounded in the GDPR and EDPB guidelines.
ProblemA breach, and the 72-hour clock is ticking - the deadline is easy to miscount.
What it doesArt. 33-34 decision tree with a deterministic 72h calculator counted from awareness. Drafts the SA notification and the communication to data subjects.
ProblemA request to access or erase data - you have one month and a few deadline traps.
What it doesClassifies the right (Art. 15-22), computes the deadline (Art. 12(3) calculator), gates exemptions and refusals, drafts the response.
ProblemRecords of processing and processor contracts - a mandatory clause is easy to miss.
What it doesBuilds the Art. 30 register and checks the processor contract against the Art. 28(3)(a)-(h) clauses, with a redline of what's missing.
ProblemA high-stakes submission goes out and you don't know where the other side will strike first.
What it doesA single-pass attack from the perspective of opposing counsel: core theory of attack, the argument stripped of rhetoric, primary lines of attack, a sceptical judge's view, surgical strikes, and what the submission is trying to hide. Cheaper than the full debate - a first filter before escalating.
ProblemYou don't know how the text actually lands on a judge reading it cold, under time pressure, with no goodwill to fill the gaps.
What it doesSimulates a sceptical first read and returns a seven-part reception report - what the case seems to be about, points of confusion, what's strong, what's weak, unproved assumptions, a calibrated confidence level, and what would be needed to persuade. It never rewrites or attacks - only reports how the text is received.
ProblemPasting case content into a cloud chatbot is a disclosure - not "tool usage" - and most people paste first and think later.
What it doesScores five risk factors (client identifiability, professional secrecy, litigation strategy, third-party data, provider terms) and returns a banded verdict - SAFE, CAUTION or STOP. On CAUTION it prepares a redacted draft for human approval; it never sends anything itself.
ProblemBefore paying for a full conformity audit, you need to know whether Regulation (EU) 2024/1689 even applies to your system - and how badly.
What it doesAn eight-question chain - AI-system definition, prohibited practices, high-risk classification, GPAI, transparency duties, operator role, obligation map with deadlines - in 20-40 minutes. Ends in a triage card for a human, not a conformity opinion.
ProblemNIS2 obligations only bite through national transposition, and member states implemented the directive differently and on different timelines - so "are we in scope" has no fixed answer.
What it doesChecks the national transposition status first, then walks sector scope, the size-cap rule, essential-vs-important classification, the ten Art. 21(2) measures, and the 24h/72h/1-month incident clock. Runs the GDPR breach clock in parallel when personal data is involved.
ProblemBefore a legal answer exists, you need to know which source controls it - and starting research at the wrong layer wastes the whole pass.
What it doesMaps an EU or member-state question onto source layers (Treaties, Charter, regulations and directives with a transposition check, CJEU case law, the national statutory chain, soft law), sets the verification order, flags collisions, and points to the right connector per layer.
Skills are open; you'll find authorship and licences in each package (_manifest.json + NOTICE). Integrity is verified by SHA256 (CHECKSUMS.txt); packages are not yet Ed25519-signed, so we don't claim a "verified" tier.
Not ours, but worth a place. Open skills and tools from other teams that hold our discipline: grounded, local, EU-aware. We link to the source with attribution and licence - we do not claim someone else's authorship. The selection grows with what we vet.
ProblemRegulation 2024/1689 is in force and your team can't tell whether a system is high-risk, GPAI, or in scope at all.
What it doesClassifies an AI system under the AI Act (prohibited, high-risk, GPAI), separates provider and deployer roles, maps the Art. 8-22 and 50 duties, generates a FRIA. Correct for the whole EU.
ProblemYou want to start with agentic legal workflows but have no vetted starting point.
What it doesAnthropic's official suite: 100+ SKILL.md procedures across 13 practice areas (M&A, privacy, IP, employment, AI governance). The method is jurisdiction-neutral - the privacy and ai-governance plugins sit closest to EU practice.
ProblemAI Act materials - FRIA questionnaires, Annex IV templates, conformity guides - are scattered across dozens of sites.
What it doesA CC0 public-domain curation of AI Act resources: templates, official sources, risk-classification tools. Raw material for your own procedures, with zero attribution friction.
ProblemPII has to be detected and masked locally before text reaches a model - regex alone isn't enough.
What it doesMicrosoft's mature PII detection and anonymisation engine (NER, patterns, checksums). Runs locally, extensible with Polish PESEL, NIP and REGON recognizers. Complements the "Let It Be" skill.
ProblemDORA, NIS2, CSRD, AI Act, ISO 42001 - each compliance framework is its own set of requirements and its own reading.
What it doesA GRC skill set with one SKILL.md per framework - the whole EU stack plus ISO and global standards. A starting point, not a replacement for your compliance officer.
ProblemA data breach, a DPIA, a privacy notice - repeatable GDPR procedures you rebuild from scratch each time.
What it doesA large, structured collection of privacy procedures (GDPR, AI Act, CCPA, LGPD and more). Take the GDPR and AI Act slice as a ready scaffold and localise it to your jurisdiction.
ProblemFor a cross-border contract you want to know which clauses are risky and how they compare to market.
What it doesAn external contract-review skill built on the CUAD taxonomy: flags risky clauses, market benchmarks, lawyer-ready redlines. Common-law native - complements our Polish clause checklist for cross-border work.
ProblemA model's contract draft reads well - you just can't tell which clause is grounded in the Civil Code and Supreme Court case law, and which one it invented.
What it doesA skill by Adam Piotrowski, radca prawny (KTZR law firm, Gdańsk): drafting and review of contracts under Polish law. A clause database, doctrine with Supreme Court case law and 12 drafting rules; plus a cross-reference check, LAWYER/LAYPERSON modes and statute verification via MCP (Sejm ELI). You install it from the author's repository.
These items belong to their authors and are offered under the licences shown (MIT, Apache-2.0, CC0). We link to the original repositories - you install them at the source; we do not host third-party packages. Attribution and the full licence live in each repository.
Some skills are built for Polish legal practice and don't translate cleanly - they are wired to Polish sources and data formats. They live on the Polish Boutique:
The English "Devil's advocate" (the red-team method, without the Polish case-law layer) is in the catalogue above. Its Polish counterpart, wired to Polish case-law, stays on the Polish Boutique.
See the Polish skills →Skills do the work on your documents. When you need statutes, case law or the company register straight in the agent - see the MCP connectors.
Skills support a lawyer's work; they don't replace their judgement. MateMatic's reading of the GDPR and the AI Act, not the position of any bar or chamber.