The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents from content libraries to extract indexing information, creating a surface for indirect prompt injection.
Ingestion points: The skill uses the Read tool to ingest the "entire content" of files from folders like 内容素材库/读书笔记库/, 核心概念库/, and 04-方法论沉淀/ to extract themes and keywords.
Boundary markers: There are no instructions to use delimiters or to treat the file content as untrusted data to avoid executing hidden instructions during the extraction process.
Capability inventory: The agent can list files (Glob), read contents (Read), write to index files (INDEX.md, 爆款索引.md), and update other skill instruction files (li-topic/SKILL.md, li-writer/SKILL.md) and the project configuration (CLAUDE.md).
Sanitization: No sanitization or filtering logic is provided for the themes, keywords, or gold sentences extracted from the source files before they are written to indexes or other configuration files.

li-index