The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by reading content from multiple local markdown files and file headers which could contain untrusted instructions.
Ingestion points: The agent reads data from 03-数据统计/数据统计表.md, 01-内容生产/选题管理/00-选题记录.md, and YAML frontmatter in 01-内容生产/待发布的内容/中短视频/.
Boundary markers: No delimiters or instructions to ignore embedded commands are used when interpolating file content into the prompt.
Capability inventory: The skill has the ability to read, edit, move, and delete files.
Sanitization: No input validation or escaping is applied to the data retrieved from the file system.
[COMMAND_EXECUTION]: The skill performs file system modifications as part of its archiving workflow.
Evidence: Step 6.2 involves moving files to a computed path in 02-已发布内容/短视频/, and Step 6.3 explicitly performs a file deletion operation in the 01-内容生产/待深化的选题/ directory based on filename matching.

li-stats