Purpose & scope. A defensive/educational knowledge base cataloguing known prompt-injection and jailbreak patterns, the models/systems they have affected, and the defenses against them. Compiled from primary literature (arXiv papers, vendor disclosures) and security research, June 2026.

How to read this. Every technique lists: how it works, an illustrative structural skeleton (the shape of the attack, not a weaponized payload), the models/systems it was reported against, and its current status. Examples are deliberately defanged. >

Cada tarea va al modelo que corresponde: opus planifica, sonnet implementa, haiku busca y lee.

1. Creá la carpeta .claude/agents/ dentro de tu proyecto (o en ~/.claude/agents/ para usarlos en todos tus proyectos).
2. Copiá ahí planner.md, coder.md y explorer.md.
3. Abrí Claude Code en ese proyecto y escribí /agents. Tienen que aparecer los 3.
4. Poné la sesión principal en Sonnet con /model sonnet, así la conversación base no gasta Opus.

This skill converts PDF files to Markdown format using Microsoft's markitdown package.

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.