The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bundled Python script scripts/article_check.py to perform quality validation on article drafts. The script calculates metrics such as word counts, paragraph distribution, and uses regex to detect banned template phrases. These operations are performed locally on user-specified files and do not involve network requests, system modifications, or elevated privileges.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from web clippings and research materials to create articles, creating a potential ingestion surface.
Ingestion points: Processes external materials stored in 30-资料/ and user-supplied draft content.
Boundary markers: Instructions explicitly mandate the use of source attribution blocks for translations and a clear distinction between source facts, author inferences, and personal experiences.
Capability inventory: File system read access via the quality check script and the ability to hand off processed content to other specialized skills (ray-cover, ray-wechat) for finishing tasks.
Sanitization: Mitigation relies on procedural fact-checking against original sources and a required human-in-the-loop review process before any content is finalized or published.

ray-writer