writing-skills

Warn

Audited by Gen Agent Trust Hub on Apr 8, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The file persuasion-principles.md teaches methods to ensure agent compliance by using 'Authority' and 'Commitment' principles. It explicitly states these techniques serve to 'eliminate rationalization' and 'override decision fatigue.' It references a paper titled 'Call Me A Jerk: Persuading AI to Comply with Objectionable Requests,' which focuses on techniques to bypass an AI's internal reasoning. These patterns (using 'YOU MUST', 'No exceptions', and 'Delete means delete') are instructional blueprints for creating highly directive prompts that aim to suppress an agent's safety-oriented judgment or ethical rationalizations.- [COMMAND_EXECUTION]: The utility script render-graphs.js uses child_process.execSync to run the system command dot. It extracts DOT language blocks from SKILL.md and pipes them into the system process. While the command itself uses a fixed string (dot -Tsvg), the script enables arbitrary system command invocation for rendering visual flows defined within documentation files.- [EXTERNAL_DOWNLOADS]: The file anthropic-best-practices.md includes several image references hosted on mintcdn.com. This domain is associated with Mintlify, a well-known documentation platform. These references are used for displaying diagrams within the technical guide and are documented here neutrally as external resource dependencies.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 8, 2026, 09:39 PM