The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process, rewrite, and translate untrusted user content, which creates a surface for indirect prompt injection attacks.
Ingestion points: The skill accepts user-provided text for '构思' (conception), '撰写' (writing), '改写' (rewriting), '翻译' (translation), and '审校' (editing) as described in SKILL.md.
Boundary markers: The instructions lack specific delimiters (like triple backticks or XML tags) to separate user content from system instructions, although it advises the agent to extract facts and constraints first.
Capability inventory: The skill is limited to text generation and transformation; no subprocess execution, file writing, or network exfiltration capabilities were found in the provided files.
Sanitization: No explicit sanitization or filtering of user-provided instructions within the data is mentioned, relying instead on the agent's internal safety filters.
[SAFE]: The skill references official Chinese national standards (GB/T 15834-2011, GB/T 15835-2011, GB/T 9704-2012) via the official openstd.samr.gov.cn domain. These are legitimate resources for a writing standards skill and do not represent a security risk.

writing-chinese