The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a specific module (persuasion-principles.md) that teaches systemic techniques to override an AI agent's reasoning and "rationalization" process. It explicitly instructs the use of "Authority" markers like "YOU MUST," "Never," and "No exceptions" to eliminate decision fatigue and bypass model objections. It further references adversarial research titled "Call Me A Jerk: Persuading AI to Comply with Objectionable Requests" to justify these methods for enforcing compliance.
[COMMAND_EXECUTION]: The utility script render-graphs.js uses child_process.execSync to execute shell commands, specifically dot (Graphviz) and which. This script extracts content from markdown files and passes it to the system shell to generate SVG diagrams, creating a surface for local command execution.
[DATA_EXPOSURE]: The skill instructions and scripts reference sensitive local directories used for storing agent logic, such as ~/.claude/skills and ~/.agents/skills/, though this is aligned with the stated purpose of managing agent capabilities.

writing-skills