The Agent Skills Directory

[PROMPT_INJECTION] (MEDIUM): The files examples/CLAUDE_MD_TESTING.md and persuasion-principles.md employ high-pressure instructional patterns such as "THIS IS EXTREMELY IMPORTANT," "If a skill existed... and you didn't use it, you failed," and "No exceptions." These tactics are specifically designed to override an agent's standard reasoning and safety decision-making in favor of rigid compliance.
[COMMAND_EXECUTION] (MEDIUM): The script render-graphs.js uses child_process.execSync to invoke the external dot (Graphviz) binary. While the script passes input via stdin rather than shell interpolation, it still enables the execution of system-level software based on the contents of markdown files.
[INDIRECT_PROMPT_INJECTION] (LOW): The utility script processes external data which could be manipulated to influence the host system through Graphviz.
Ingestion points: Reads SKILL.md files from user-defined directories using fs.readFileSync.
Boundary markers: Uses markdown code blocks (```dot) as markers to extract content.
Capability inventory: Performs file writes (fs.writeFileSync), directory creation (fs.mkdirSync), and system command execution (execSync).
Sanitization: There is no validation or sanitization of the extracted DOT content before it is passed to the system binary.

writing-skills