The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by ingesting untrusted data from XiaoHongShu notes that the agent is then tasked to analyze.\n
Ingestion points: The scripts/fetch_note_content.ts script fetches titles, descriptions, and tags from external XHS URLs.\n
Boundary markers: There are no explicit delimiters or instructions to the agent to ignore instructions embedded within the scraped content.\n
Capability inventory: The skill can execute shell commands via curl, write files to the workspace, and perform browser automation using camoufox-cli.\n
Sanitization: No sanitization or filtering of the extracted text content is performed before analysis.\n- [EXTERNAL_DOWNLOADS]: The skill downloads images and metadata from xiaohongshu.com and xhslink.com using curl and fetch. This is a legitimate requirement for the skill's content research functionality.\n
Evidence: Network operations in scripts/fetch_note_content.ts target XiaoHongShu domains for asset retrieval.\n- [COMMAND_EXECUTION]: The skill invokes curl using child_process.execFile to resolve short links and download images. This method uses an array of arguments, which effectively mitigates common shell injection risks compared to string-based command execution.\n
Evidence: execFileAsync("curl", [...]) calls are used for redirect resolution and image downloading in scripts/fetch_note_content.ts.

xhs-content-ops