Claude Skills by valorisa

This skill transforms the paper *ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?* into structured, reusable security artifacts. The paper introduces a benchmark for evaluating AI agents on real-world exploitation tasks across userspace, V8, and Linux kernel targets, with 898 instances and containerized evaluation environments.

Orchestration d'une revue technique collaborative multi-étapes destinée à transformer une analyse initiale en un artefact technique robuste (documentation, article, dépôt GitHub, spécification ou rapport).

Structure any multi-step, iterative task on a GitHub repo (fixing CI, resolving lint violations, iterating on a PR, debugging a failing workflow) as an observe-decide-execute-verify loop with explicit stop conditions and guardrails. Use this whenever a task could require more than one attempt to succeed — CI is failing and the fix isn't obvious on the first try, a task says "keep trying until X passes", or you're about to loop on the same file/branch multiple times. Do NOT use for single-shot...

Moteur de pensée divergente. Retourne le cadrage évident d'un sujet ou d'une décision pour faire surgir les angles qu'on ne voit pas. Ose beaucoup, ne valide rien — chaque angle sort tagué. À mobiliser dès que l'utilisateur tourne en rond sur une seule façon de voir, cherche un angle neuf pour une vidéo, hésite sur une décision, ou demande explicitement "angle mort", "retourne ça", "angle inverse", "contre-pied", "quel angle on rate", "vois ce que je rate", "fais péter mes angles morts". Util...

Enforce rigorous factual accuracy, epistemic discipline, source integrity, uncertainty handling, and resistance to uncritical agreement. Prevent fabricated information, unsupported certainty, and automatic acceptance of user premises.

Conceives the shortest path to personalized net revenue by analyzing profile, audience, offer, funnel, unit economics, and conversion bottlenecks. Use when the user wants a monetization strategy, offer architecture, funnel diagnosis, or a practical path to high-margin revenue.

Skill d'agent qui impose le décodage contraint pour toutes les sorties destinées à être lues par du code (agents, outils, API, BDD). Elle sépare explicitement la phase de raisonnement libre de la phase de sortie structurée conforme à un schéma ou une grammaire.

Break a plan, spec, or PRD into independently-grabbable GitHub issues using tracer-bullet vertical slices. Use when user wants to convert a plan into issues, create implementation tickets, or break down work into GitHub issues.

Orchestration d'un dialogue structuré entre deux LLMs (A et B) autour d'un même sujet, avec cadrage d'intention, génération d'angles morts, recherche systématique de contradictions et garde-fous contre le contournement silencieux.

Discipline de travail haut de gamme à tenir toute la session. Déclencher dès que l'on veut un travail soigné, vérifié, honnête : 'fabuleux', 'mode fabuleux', 'pense comme Fable', 'niveau Fable 5', 'qualité maximale', 'sois exigeant', 'relis vraiment'. ROUTE selon le type de tâche : ARTEFACT/AGENTIQUE (page, deck, code, doc, données) → produire + screenshot + vision + correction ; PROSE (email, post, article) → draft + passe de soustraction ; ANALYSE/CONSEIL → critères + vérification de chaque...

Analyze a file to detect bugs and propose fixes.

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

Interviewer l'utilisateur sans relâche sur un plan ou un design jusqu'à compréhension mutuelle complète, en résolvant chaque branche de l'arbre de décision. À utiliser pour stress-test un plan, se faire griller sur un design, ou sur mention "grille-moi".

Skill de protection contre le contournement silencieux, les faux succès et les dérives d'interprétation dans les systèmes agentiques. Elle aide à formaliser l'intention, vérifier les hypothèses et imposer des critères d'arrêt clairs.

Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Based on Karpathy's LLM Council methodology. MANDATORY TRIGGERS: 'council this', 'run the council', 'war room this', 'pressure-test this', 'stress-test this', 'debate this'. STRONG TRIGGERS (use when combined with a real decision or tradeoff): 'should I X or Y', 'which option', 'what would you do', 'is this the right move', 'v...

Optimise l'usage d'un long contexte d'entrée en structurant explicitement le placement des morceaux de contexte (chunks) en début, milieu et fin, pour atténuer l'effet "Lost in the Middle" mis en évidence par Liu et al. (2023).

Un workflow multi-agents avancé conçu pour résoudre, vérifier et mettre en forme des preuves rigoureuses pour les compétitions de mathématiques (telles que l'IMO, Putnam, USAMO, AIME). Il s'appuie sur des agents de résolution parallèles, une vérification adversariale avec isolation de contexte, un "mode profond" (deep mode) pour le calcul borné, et une passe de présentation finale pour générer des preuves élégantes en LaTeX. L'objectif est d'optimiser la précision conditionnelle (ne proposer ...

Organise un débat structuré entre plusieurs LLM sur un plan, un design ou une idée de projet, avec un cadrage silencieux de type "grill me", des rounds de critique croisée, puis un juge final. Utilise cette skill dès qu’un plan, une architecture, un design ou une idée de projet doit être stress-testé par plusieurs modèles avant décision.

Monitors task execution for skill improvement opportunities. Use this skill during ANY multi-step task, agentic workflow, or substantive work session where the agent is using tools and producing deliverables. It captures patterns, user corrections, workflow insights, and methodology worth preserving as reusable skills. Also triggers during post-task feedback discussions and when the user explicitly mentions skill observations, improvements, the observation log, skill taxonomy, or asks the age...

> **Auditable pipeline with optional multi-agent validation**: Generate custom prompts via 5 traced validation circles + 18 fused hacks. Council option for external audit by 5 independent advisors with blind peer review (Karpathy methodology). ---

Generate optimized, domain-agnostic prompts via a 5-circle validation pipeline fused with 18 optimization hacks. Produces auditable, copy-paste-ready prompts for any AI tool. TRIGGERS: 'create a prompt', 'optimize this prompt', 'promptor', '/promptor', 'generate a system prompt', 'prompt engineering', 'build me a prompt for'. Also triggers on: 'reverse prompt engineer this', 'improve this prompt', 'prompt audit'.

Use this skill when the user asks to "audit repo", "analyze repository", "find bugs", "find gaps", "review codebase", "check code quality", or wants to analyze a repository for bugs, missing features, test coverage gaps, security issues and create GitHub issues for findings. Performs a deep technical audit and optionally opens GitHub issues for high-priority problems.

Use when user hits rate limits, quota exceeded errors, context exceeds 40%, or exhibits token waste symptoms (long conversations, wrong model, verbose output, MCP bloat, cache invalidation, expensive files, too many requests)

Use when user wants deep intellectual debate on societal topics (politics, economics, philosophy), asks for Socratic discussion, or needs anti-echo-chamber sparring partner who will challenge their positions rather than validate them

Create new skills for Claude Code. A skill is just a markdown file — nothing more.

Generate a complete REST API endpoint from a one-line description.

Advanced meta-skill for creating, reviewing, validating, and improving Claude skills following Anthropic's official patterns. Use when the user asks to build a production-ready skill, audit skill quality, generate complete skill packages with tests and validation, review skill architecture, or optimize skill triggering. Handles all 3 categories (document creation, workflow automation, MCP enhancement).

Meta-skill for skill lifecycle management. Monitors active skills, optimizes context window by unloading stale skills, detects dependency conflicts, recommends skills based on project type, performs health checks on MCP servers and APIs, and archives session memory for efficient future loading. Use when user says "optimize skills", "clean up context", "manage skills", "free tokens", "skill health check", "session feels slow", or when context >60% and multiple skills loaded. Also auto-triggers...

Activate spec-driven development mode with enforced pipeline (SPEC→PLAN→IMPL→VERIF→SYNTHESE), 3-way triage (FULL/LIGHT/SHIP), token budgets, and explicit gates. Use when starting a feature, refactoring, or complex task that benefits from structured spec-first workflow. TRIGGERS: 'spec-driven', '/spec-driven', 'mode spec', 'spec first', 'pipeline complet', 'workflow structure'. Do NOT trigger on simple questions, quick fixes, or when user explicitly wants fast/informal mode.

Automatic GitHub repo standardization at publication checkpoints with bilingual READMEs (EN/FR), CI linting, badges, and About configuration

À utiliser quand on demande de lancer une Storm Recherche, d'utiliser le skill storm-research, d'appliquer la méthode STORM à un sujet, dit « storm recherche sur X » / « briefing STORM sur X ». Ce skill exécute la méthode STORM en plusieurs phases et produit un briefing HTML vérifié.

Consigne frictions et apprentissages en cours de tâche.

Test-driven development combining strict discipline (Iron Law, mandatory verification) with intelligent workflow (planning, vertical slicing, domain awareness). Use when implementing features or fixing bugs with TDD.

Optimize token consumption in Claude Code through cache management, context forking, model selection, and input filtering. Use when user mentions high token costs, context window bloat, slow sessions, cache invalidation, verbose agent output, heavy MCP load, or any context management issues. Covers four critical axes - cache misses, context bloating, wrong model/effort level, and verbose input formats. Triggers on "tokens", "cost", "expensive", "context", "cache", "slow session", "optimize", ...