paper-write

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists of Markdown-based prompt templates and documentation. It does not contain any scripts (Python, Node.js), executables, or configurations that would lead to code execution or system modification.
  • [DATA_EXFILTRATION]: No network access commands (e.g., curl, wget) or hardcoded credentials were identified. The skill processes user-provided text locally within the agent's context.
  • [PROMPT_INJECTION]: The skill includes instructions to bypass AI detection ('anti-aigc' or '去 AI 痕'). While these are constraints for writing style, they are not directed at overriding the underlying LLM's safety filters or extracting system instructions.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process untrusted data provided by the user (theses drafts, reference papers). This represents a standard attack surface for indirect prompt injection.
  • Ingestion points: User-provided 'drafts', 'sample papers', and 'outlines' across multiple reference files (e.g., structure-imitate-general.md).
  • Boundary markers: The prompts utilize brackets and placeholders (e.g., 【中文草稿】, [在此处粘贴]) to separate instructions from user data.
  • Capability inventory: The skill performs text generation and can interact with another skill (pptgen-drawio) to generate PPT files.
  • Sanitization: There is no evidence of input sanitization or filtering of instructions embedded within the user-provided text.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:32 AM