avoid-ai-writing
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill's measurement and corpus management scripts fetch datasets and human-authored reference text from well-known and reputable services including HuggingFace (huggingface.co), Project Gutenberg (gutenberg.org), the IETF (rfc-editor.org), and the W3C (lists.w3.org). These downloads are restricted to text-based data for benchmarking the detection engine and do not involve script execution.
- [COMMAND_EXECUTION]: The skill includes several Node.js utility scripts for quote normalization, style checking, and running the detection engine. These are intended to be executed by the agent to process local files during rewrite or audit tasks.
- [INDIRECT_PROMPT_INJECTION]: As a text processing tool, the skill ingests untrusted user content. It contains explicit defensive instructions in its system prompt (SKILL.md) to treat ingested text strictly as data and to ignore any commands or behavioral overrides embedded within the analyzed content.
- [OBFUSCATION]: The detection engine includes logic to identify and strip obfuscation techniques (such as zero-width characters and homoglyphs) from user input to prevent bypasses. The skill's own implementation is transparent and does not use obfuscation.
Audit Metadata