The Agent Skills Directory

[PROMPT_INJECTION] (MEDIUM): The 'persuasion-principles.md' file provides a detailed framework for using 'Authority' and 'Commitment' principles to ensure AI compliance by bypassing its reasoning. It recommends specific linguistic patterns such as 'YOU MUST', 'Always', and 'No exceptions', which are primary markers of prompt injection intended to override agent defaults.\n- [COMMAND_EXECUTION] (MEDIUM): The 'render-graphs.js' script uses 'child_process.execSync' to invoke the system 'dot' command. Although the command string itself is static, it processes graph data extracted directly from 'SKILL.md' files, which constitutes a risk if the input is maliciously crafted or if the external utility contains vulnerabilities.\n- [PROMPT_INJECTION] (LOW): 'CLAUDE_MD_TESTING.md' includes test scenarios that utilize emphatic markers and directive instructions (e.g., 'THIS IS EXTREMELY IMPORTANT', 'If a skill existed... and you didn't use it, you failed') designed to pressure agents and override standard decision-making processes.

writing-skills