The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The script render-graphs.js uses child_process.execSync to call the system dot binary. While it passes data via standard input rather than shell arguments, executing external binaries based on file content (the SKILL.md dot blocks) remains a security surface.
[PROMPT_INJECTION] (MEDIUM): The files persuasion-principles.md and CLAUDE_MD_TESTING.md explicitly document and encourage the use of 'Authority' and 'Commitment' principles (e.g., using 'YOU MUST', 'No exceptions', 'Announce skill usage') to force AI compliance. These techniques are characteristic of prompt injection and are used to bypass an agent's autonomous reasoning or safety guidelines under the guise of 'discipline-enforcing' documentation.
[DATA_EXPOSURE] (SAFE): The provided files do not contain hardcoded credentials or access sensitive user paths (like ~/.ssh). The script only reads the specific SKILL.md provided via arguments and writes to a diagrams subdirectory.
[EXTERNAL_DOWNLOADS] (SAFE): No network operations (curl, wget, fetch) were detected. The script requires graphviz to be pre-installed on the host system.

writing-skills