The Agent Skills Directory

[PROMPT_INJECTION]: The persuasion-principles.md file explicitly advocates for the use of "Authority" and "Commitment" principles to bypass agent "decision fatigue and rationalization." It instructs the use of imperative language like "YOU MUST," "Never," and "No exceptions" to enforce compliance, which are techniques commonly used to override an agent's internal reasoning or safety filters.
[COMMAND_EXECUTION]: The render-graphs.js script uses execSync to call the system command dot. It extracts Graphviz DSL content from SKILL.md and passes it as input to the dot utility. Executing system binaries on content extracted from markdown files poses a risk if the file content is maliciously crafted.
[DYNAMIC_EXECUTION]: The skill includes a Node.js utility, render-graphs.js, which dynamically extracts and compiles Graphviz code at runtime to generate SVG files. This involves executing external processes based on document content.
[METADATA_POISONING]: The persuasion-principles.md file cites a research paper, "Meincke et al. (2025)," describing large-scale AI compliance testing. This appears to be a future-dated or fabricated reference used to provide a false sense of authority and legitimacy to the psychological manipulation techniques recommended in the skill.

writing-skills