The Agent Skills Directory

[SAFE]: The skill operates under a strict least-privilege model, restricting browser capabilities to specific browser.* facades and limiting interaction to the authorized domains xiaohongshu.com and creator.xiaohongshu.com.
[SAFE]: The skill includes explicit security instructions to treat all browser snapshots as untrusted data, preventing indirect prompt injection by ensuring the agent does not interpret web content as instructions.
[SAFE]: A mandatory human-in-the-loop requirement is enforced for all sensitive operations (publishing, scheduling, commenting, account switching), requiring an explicit chat preview and user confirmation before execution.
[SAFE]: The skill utilizes a local, isolated Python helper script (xhs_contract.py) to validate and parse data snapshots deterministically, which mitigates the risk of the LLM misinterpreting or being manipulated by malicious page structures.
[SAFE]: The skill design explicitly prohibits the extraction, clearing, or transmission of sensitive account data such as session cookies, protecting the user's account integrity.

xiaohongshu