The Agent Skills Directory

PROMPT_INJECTION (MEDIUM): The files persuasion-principles.md and CLAUDE_MD_TESTING.md contain instructions designed to override agent behavior using 'Authority' and 'Commitment' principles. Examples include 'YOU MUST', 'No exceptions', and 'If a skill existed... and you didn't use it, you failed'. These patterns are intended to bypass the agent's natural decision-making process.
COMMAND_EXECUTION (MEDIUM): The script render-graphs.js uses child_process.execSync to call the system command dot. While it uses the input option to pipe data to stdin, it still constitutes dynamic command execution based on content parsed from SKILL.md.
INDIRECT_PROMPT_INJECTION (LOW): The render-graphs.js script ingests untrusted data from SKILL.md and passes it to a subprocess.
Ingestion points: SKILL.md (via fs.readFileSync in render-graphs.js).
Boundary markers: None detected for the DOT content extraction.
Capability inventory: execSync for system commands and fs.writeFileSync for file operations.
Sanitization: No sanitization is performed on the extracted DOT blocks before they are passed to the dot command.

writing-skills