The Agent Skills Directory

[COMMAND_EXECUTION]: The render-graphs.js script uses child_process.execSync to run the dot utility and the which command. This allows the skill to execute subprocesses on the underlying operating system.
[PROMPT_INJECTION]: The skill instructions frequently use imperative and absolute language (e.g., 'YOU MUST', 'No exceptions', 'Delete means delete') designed to override the agent's internal reasoning and safety filters. The file persuasion-principles.md explicitly advocates for using psychological 'Authority' and 'Commitment' principles to increase compliance rates.
[PROMPT_INJECTION]: The testing methodology in testing-skills-with-subagents.md utilizes role-playing scenarios and pressure markers such as 'IMPORTANT: This is a real scenario', which are functionally identical to common jailbreak patterns.
[COMMAND_EXECUTION]: The render-graphs.js utility processes external input from SKILL.md files and feeds it into the dot system binary without sanitization, creating a surface for potential command-related exploits if the underlying binary is vulnerable.

writing-skills