The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The script render-graphs.js executes system commands to render diagrams. * Evidence: Line 79 uses execSync to run the dot binary. * Evidence: Line 110 uses execSync to run which dot. * Risk: Interaction with system binaries outside the agent's restricted environment.
PROMPT_INJECTION (MEDIUM): Documentation details methods to ensure compliance and prevent agent rationalization using authority and scarcity. * Evidence: persuasion-principles.md recommends using 'YOU MUST' and 'No exceptions'. * Evidence: CLAUDE_MD_TESTING.md uses emphatic blocks like '<important_info_about_skills>' to force skill usage and discourage independent reasoning. * Risk: These instructions are designed to override an agent's safety protocols and judgment by creating non-negotiable compliance paths.

writing-skills