The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a feature for publishing to WeChat via a "custom interface" that defaults to a third-party IP address (http://39.108.254.228:8002/publish-draft). Using this feature involves sending generated article content and metadata to an unverified external server not associated with a trusted service, which poses a data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from Zhihu hot lists, Weibo, and RSS feeds via scripts/fetch_sources.py to drive content generation. Analysis confirms that there are no boundary markers or sanitization logic applied to this external content. Maliciously crafted information within these hot topics could influence the agent's output or trigger unintended tool usage. Capabilities involved include network operations via requests.post to publishing platforms and local file system interactions.
[COMMAND_EXECUTION]: The script scripts/auto_publisher.py utilizes subprocess.run to execute other scripts within the package, and scripts/copy_to_clipboard.py uses subprocess.Popen to interact with system clipboard utilities like xclip or xsel. While these operations are currently configured for standard functionality, they represent a surface for local command execution.

wechat-hotspot-publisher