writing-skills
Warn
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation ('persuasion-principles.md' and 'testing-skills-with-subagents.md') instructs agents on how to use psychological manipulation techniques—including Authority, Commitment, and Scarcity—to ensure compliance with instructions. It explicitly utilizes override markers like 'IMPORTANT: This is a real scenario' and references research on persuading AI to comply with objectionable requests to bypass an agent's internal reasoning or 'rationalization' during automated testing scenarios.\n- [COMMAND_EXECUTION]: The 'render-graphs.js' script uses the Node.js 'child_process.execSync' function to execute the 'dot' command-line utility. The script takes input directly from markdown files ('dot' blocks) without sanitization, which presents a risk of command injection if an agent is tasked with processing untrusted or malicious skill files.
Audit Metadata