Djordjes-Claude-Setup

Evidence-based dialogue mode that replaces sycophantic AI responses with structured critical analysis. Use when the user presents a claim, idea, plan, or argument and needs genuine critical engagement — not validation. Activate on phrases like "what do you think about...", "is this right?", "am I thinking about this correctly?", "challenge this", "steelman/critique this", or any time genuine intellectual pushback would serve the user better than agreement. Five modes: FULL (4-move deep analysis), INTERACTIVE (Socratic step-by-step), TLDR (3-5 line quick take), STEELMAN (strongest case for + against), DECISION (tradeoff table + verdict).

Use this skill whenever the user is building something with the Claude API and would benefit from a known-good reference implementation from Anthropic's official Claude Cookbooks. Trigger on any request that touches Claude API features — RAG, tool use, agents (Agent SDK or Managed Agents), vision/multimodal, prompt caching, batch processing, extended thinking, citations, JSON mode, message batches, finetuning, evals, skills, memory, parallel tools, structured output, sub-agents — or on integrations (Pinecone, MongoDB, LlamaIndex, Wikipedia, Wolfram, ElevenLabs, Deepgram, VoyageAI). Trigger even when the user doesn't explicitly say "cookbook": questions like "how do I do X with Claude", "what's the recommended pattern for Y", "show me how to build Z with the Anthropic API", "I want to add tool use to my app", or "best way to handle long contexts" should pull this skill. Do NOT trigger for unrelated coding questions, Claude.ai UI questions, or pure prompting questions that don't involve the API.

Convenes 7 expert AI personas to debate any decision, idea, or problem and synthesizes a structured verdict with confidence score, minority report, and concrete next steps. Use when the user presents a decision, asks "should I...", "is this a good idea", "help me decide between...", "stress-test this idea", or says "convene the council". Automatically triggers on any genuine tradeoff with stakes. Produces genuine disagreement — not a balanced summary — with a clear verdict and no "it depends" without specifics.

Structured decision-making frameworks for complex, high-stakes, or ambiguous choices. Use when the user needs to make a significant decision and would benefit from a formal framework — business strategy, tech architecture, career moves, startup pivots, build-vs-buy, prioritization. Activate on: "how should I decide", "help me think through", "what framework", "prioritize these", "should we build or buy", "is this worth it". Provides: RICE scoring, decision matrices, pre-mortem analysis, second-order thinking prompts, and synthesis into a clear recommendation.

Create interactive HTML learning tools — flashcard decks, visual theory explainers, and concept pages — for studying. Use this skill whenever the user wants to learn, review, or practice a topic interactively, wants flashcards, a visual summary, a quiz, or says things like "erkläre mir X interaktiv", "mach mir Lernkarten zu X", "ich will X üben", "help me study X", "make a study page for X", or "build me a flashcard deck on X". Covers all subjects including mathematics, statistics, programming, general education (BM/Schule), and quantitative finance/trading. Always prefer this skill over plain text explanations when the user wants to actively learn or review something, not just read about it.

Analysiert Kryptowährungen auf Rugpull-Risiken und Betrugszeichen. Verwende diesen Skill IMMER wenn: - Ein Nutzer fragt ob ein Coin/Token sicher ist, einen Rugpull-Score will, oder einen Coin analysieren möchte - Jemand einen Coin-Namen + Social Media Erwähnung nennt (z.B. "FOFAR wird von timclassified beworben") - Jemand fragt "Ist [Coin] ein Scam?", "Lohnt sich [Token]?", "Ist das ein Rugpull?" - Krypto-Coins mit verdächtigen Merkmalen erwähnt werden: unbekannte Tokens, Meme-Coins, neue Launches - On-Chain Analyse, Tokenomics-Check oder Influencer-Werbung für Kryptos untersucht werden soll Gibt einen detaillierten Rugpull-Report mit Score 0-100 und Ampelsystem zurück. Nutze diesen Skill auch proaktiv wenn du Betrugszeichen erkennst, auch wenn nicht explizit nach einem Score gefragt wird.

Converts any documentation website, GitHub repo, API reference, or technical resource into a ready-to-upload Claude skill (SKILL.md file). Use when the user wants to turn documentation into a persistent skill — e.g. "make a skill from the FastAPI docs", "convert the Anthropic API reference into a skill", "I want Claude to know the Supabase docs", "turn this GitHub README into a skill". Analyzes the source, extracts the most useful patterns and procedures, and outputs a complete SKILL.md file.

Apply Clean Architecture, SOLID principles, design patterns, and software architecture best practices when designing systems, reviewing code structure, planning a new project, or refactoring existing code. Use when the user asks about architecture decisions, how to structure a project, which design pattern to use, how to apply SOLID, or when reviewing code for architectural quality. Covers: Clean Architecture layers, SOLID, DDD, common patterns (Repository, Factory, Strategy, Observer, CQRS), and anti-patterns to avoid.

Aktiviere diesen Skill IMMER wenn der User /ultraplan schreibt oder eine Aufgabe kommt die eine tiefe, strukturierte Analyse erfordert. Behandelt jede Aufgabe wie eine Militäroperation: Lageerkundung (Web Research), Lagebeurteilung (Analyse mit vollem Kontext über den User), Operationsplan (klare Empfehlung), Ausführung (präzise Antwort). Kein Raten, kein Bullshit, kein unnötiger Output. Trigger auch bei: "analysier das für mich", "was ist besser", "wie soll ich vorgehen", "mach einen Plan", "deep dive", "ultraplan".