The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Node.js utility script, render-graphs.js, which executes the system command dot (Graphviz) via execSync. This is used to transform dot code blocks within SKILL.md into SVG diagrams for visualization. The script uses standard input for the graph data and sanitizes output filenames using a alphanumeric regex, which mitigates common shell injection and path traversal risks.
[PROMPT_INJECTION]: The skill documentation, particularly in persuasion-principles.md, instructs the AI to use 'Authority' and 'Commitment' principles (e.g., 'YOU MUST', 'No exceptions', 'Delete means delete') to ensure compliance with the TDD methodology. While these are powerful directives designed to override an agent's tendency to rationalize skipping steps, they are used to enforce the skill's own quality standards rather than to bypass safety filters or extract system information.
[DATA_EXPOSURE]: There are no hardcoded credentials, secrets, or references to sensitive local files (such as SSH keys or environment configurations) within the skill files.
[EXTERNAL_DOWNLOADS]: The skill does not perform unauthorized network requests or download external code. It includes a copy of Anthropic's official best practices for context, which is a trusted source.

writing-skills