The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from markdown archives in the project's session directory.
Ingestion points: Reads multiple .md files from the {存档根目录}/sessions/{项目名}/ path.
Boundary markers: The instructions include explicit constraints such as "搬运，加索引，仅此而已" (Carrying, adding index, that is all) and "不从对话里凭空补内容" (Do not supplement content from conversation), which act as boundaries to prevent the agent from executing instructions found in the files.
Capability inventory: The skill utilizes file system access to read archives, write reports, and create directories using mkdir -p.
Sanitization: The skill validates the presence of specific headers and sections in the input files but does not sanitize the text content, as it is designed for verbatim data transcription.
[COMMAND_EXECUTION]: The skill performs administrative filesystem tasks to organize reports. It creates a reports/ directory and writes new files using paths derived from project configuration. These operations are scoped to a specific project root and represent the skill's intended functionality.

xy-brief