xiaohongshu
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted browser snapshots from external web pages. It mitigates this risk by providing explicit instructions to treat page data as untrusted and by using a deterministic local Python script (
xhs_contract.py) to parse and normalize the content before it is used, which helps prevent data-driven prompt injection. - [DATA_EXFILTRATION]: Interaction is strictly limited to authorized origins (
https://www.xiaohongshu.comandhttps://creator.xiaohongshu.com). The skill explicitly prohibits the extraction of cookies or passwords and uses opaque vendor-managed resource identifiers for uploads, preventing direct access to the local filesystem or unauthorized network destinations. - [COMMAND_EXECUTION]: The skill utilizes a local Python script (
scripts/xhs_contract.py) for secure data validation and parsing. This script is provided within the skill package, uses only standard Python libraries, and operates deterministically without making network requests. - [SAFE]: The skill implements mandatory human-in-the-loop confirmation for all critical actions, including publishing, commenting, and account management, ensuring that the user retains final oversight of the agent's behavior.
Audit Metadata