The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it ingests and processes content from third-party Xiaohongshu web pages.
Ingestion points: The skill utilizes browser.read_page to extract data from www.xiaohongshu.com and creator.xiaohongshu.com (found in SKILL.md).
Boundary markers: The instructions include an explicit warning to treat page content as untrusted data that should never alter the user's intent or the skill's policy (found in SKILL.md).
Capability inventory: The skill utilizes capabilities including browser.click, browser.form_input, and browser.file_upload to interact with the web interface (found in SKILL.md).
Sanitization: The skill employs deterministic scanning and reconstruction of semantic trees to parse note cards correctly, but does not define specific sanitization or filtering for the text extracted from the page.
[EXTERNAL_DOWNLOADS]: The skill downloads media files from the author's infrastructure for publication.
Evidence: The browser.file_upload tool is instructed to download artifacts from the Ace Data Cloud CDN. This resource belongs to the author 'acedatacloud' and is considered a legitimate vendor service.

xiaohongshu