The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a Python script (archive_rednote.py) to automate content retrieval and file management. This script writes Markdown notes to the Obsidian vault and downloads media binaries (images and videos) to the ~/Downloads/soia-pkm-clip-rednote/ directory.
[DATA_EXFILTRATION]: The script performs network operations using urllib.request to fetch data from xiaohongshu.com and xhslink.com. It includes a proactive security control (validate_host) that verifies the target domain before making requests, specifically to ensure that the XHS_COOKIE (if configured) is not transmitted to unauthorized third-party hosts.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by processing untrusted data from external Xiaohongshu posts.
Ingestion points: archive_rednote.py scrapes the title and desc (description) fields from the target Xiaohongshu URL.
Boundary markers: The scraped content is written to a Markdown file under the ## 原文 header. There are no specialized delimiters or system instructions to the agent to ignore embedded instructions within this block.
Capability inventory: The skill performs file writes and network requests. The agent is explicitly instructed in SKILL.md to read the resulting file and generate a summary, which creates a vector for instruction obedience from the untrusted data.
Sanitization: Filenames are sanitized, but the description text is written to the vault without escaping or filtering for malicious instructions.

soia-pkm-clip-rednote