The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/clip_zhihu_official.py makes network requests to developer.zhihu.com and api.tikhub.io. TikHub is used to fetch full-text content when the official API provides only summaries. This is an intended feature of the skill as documented in the README.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection (Category 8) due to its core function of ingesting and processing untrusted web content.
Ingestion points: Data enters via API calls in scripts/clip_zhihu_official.py fetching content from Zhihu and TikHub.
Boundary markers: The skill uses YAML frontmatter and Markdown headings to structure output, though it does not explicitly warn the LLM to ignore instructions inside the clipped content.
Capability inventory: The skill can write to the local file system and perform network requests for content retrieval and heading review.
Sanitization: Content is sanitized through MarkdownHTMLParser and regex-based cleanup routines, reducing the risk of embedded malicious code execution during rendering.
[SAFE]: The script reads API keys from environment variables or the Windows Registry (HKEY_CURRENT_USER\Environment). This is a legitimate and common practice for CLI tools to handle authentication without hardcoding secrets.
[SAFE]: The skill provides an optional heading review feature that sends article outlines to an OpenAI-compatible API. The script only sends extracted metadata (headings) and applies numeric level adjustments, which does not involve dynamic code execution.

zhihu-clippings-vp