The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The file render-graphs.js utilizes child_process.execSync to invoke the system dot binary. While the command string is static and input is passed via stdin, executing system-level binaries based on content parsed from markdown files (SKILL.md) presents a risk of exploiting potential vulnerabilities in the underlying rendering engine (Graphviz).
[PROMPT_INJECTION] (LOW): The files persuasion-principles.md and examples/CLAUDE_MD_TESTING.md describe and test techniques designed to force agent compliance. These include using 'Authority' patterns (e.g., "YOU MUST", "No exceptions") and XML-like structural markers to override an agent's default decision-making process. While intended for skill adherence, these patterns mirror adversarial prompt injection techniques used to bypass standard safety or logic constraints.

writing-skills