The Agent Skills Directory

[PROMPT_INJECTION]: The skill and its supporting documentation (persuasion-principles.md, testing-skills-with-subagents.md) explicitly teach and model the use of psychological persuasion techniques to override the agent's internal reasoning and 'rationalization' processes. It instructs the use of high-pressure language (e.g., 'YOU MUST', 'No exceptions', 'Violating the letter... is violating the spirit') and cites research on persuading AI to comply with 'objectionable requests' as a foundation for enforcing discipline.
[COMMAND_EXECUTION]: The included utility script render-graphs.js uses child_process.execSync to execute the system dot (Graphviz) command and which dot. While functionally necessary for its stated purpose, it represents arbitrary shell command execution triggered by processing markdown content.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to use various shell commands such as ls, grep, wc, and rm (implied by 'delete means delete') to manage and verify skill files in the user's directory.
[PROMPT_INJECTION]: The skill includes 'Pressure Scenarios' designed to manipulate the agent into choosing specific behaviors by simulating high-stress environments, such as production outages costing $10k/minute, to test whether the agent will bypass safety or procedural guidelines.

writing-skills