The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains extensive documentation (e.g., persuasion-principles.md, testing-skills-with-subagents.md) on how to use 'Authority', 'Commitment', and 'Scarcity' principles to override an agent's reasoning. It specifically instructs how to 'bulletproof' skills against agent 'rationalization' using imperative language (e.g., 'YOU MUST', 'No exceptions') to ensure compliance even under pressure. While framed as a method for documentation quality, these are the fundamental techniques used in behavior-override and jailbreak scenarios.
[COMMAND_EXECUTION]: The render-graphs.js script uses the child_process.execSync module to execute the system's dot (Graphviz) command. It processes data extracted from the skill's markdown files and passes it to the external process. Although it uses standard input for the data transfer, the execution of external binaries based on file content carries inherent risks if the processing tool or the data parsing has vulnerabilities.

writing-skills