The Agent Skills Directory

[PROMPT_INJECTION]: The files persuasion-principles.md and testing-skills-with-subagents.md contain instructions that mirror prompt injection and jailbreak techniques. They advocate for the use of "Authority" markers (e.g., "YOU MUST", "Never", "No exceptions") and "Pressure scenarios" (e.g., "IMPORTANT: This is a real scenario. Choose and act") to override an agent's reasoning or "rationalizations." These patterns are designed to force compliance by using DAN-style framing to simulate high-stakes environments and bypass an agent's natural judgment.
[COMMAND_EXECUTION]: The render-graphs.js script utilizes child_process.execSync to invoke the dot command-line utility from the Graphviz suite to render SVG diagrams. While the command itself is static, it interacts directly with the host system's shell environment.
[DYNAMIC_EXECUTION]: The render-graphs.js script dynamically extracts dot language code blocks from SKILL.md using regular expressions and pipes them into the dot system process via stdin. This creates a bridge where content from a markdown documentation file is processed as executable input for an external rendering engine.

writing-skills