xiaoqi-style-writer
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted user input for the purpose of writing style extraction, creating a surface for indirect prompt injection.
- Ingestion points: User-provided text is ingested in
SKILL.mdunder the '输入模式 C — 用户提供文本' (Input Mode C — User Provided Text) section. - Boundary markers: The instructions lack explicit text delimiters (like XML tags or triple quotes) when instructing the agent to process the user input, though it does specify a 'silent execution' of the extraction engine.
- Capability inventory: The agent has the capability to read and write markdown/JSON files within the
styles/andmy-styles/directories to manage the style library. - Sanitization:
EXTRACTION_ENGINE.mdcontains strong sanitization guidelines, specifically the '污染检查' (Pollution Check) step, which forces the agent to remove all content-bound elements like names, places, and specific references, focusing only on structural 'bone' features. This significantly mitigates the impact of instructions embedded within the processed text.
Audit Metadata