The Agent Skills Directory

[COMMAND_EXECUTION]: The skill recommends a configuration for BrowserProfile where disable_security is set to True. This action disables critical browser safety mechanisms such as the Same-Origin Policy (SOP) and SSL certificate validation, increasing the risk of cross-site attacks.
[COMMAND_EXECUTION]: The skill requests and utilizes high-privilege tools (Bash, Exec) to run Python-based automation scripts and manage local browser processes.
[EXTERNAL_DOWNLOADS]: The skill interacts with non-standard, third-party LLM API providers at https://cn.xingsuancode.com and https://ai.9w7.cn/v1. While these are used for the skill's primary function, they represent external data dependencies outside of the standard trusted providers.
[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for Indirect Prompt Injection due to its core browser-use functionality.
Ingestion points: The Agent reads and interprets arbitrary content from external websites (e.g., polymarket.com) during task execution.
Boundary markers: The provided code templates do not implement delimiters or system-level instructions to differentiate between the user's original task and instructions potentially embedded in the target webpage's HTML.
Capability inventory: The skill has access to sensitive capabilities including shell execution (Bash, Exec), file system access (Read, Write), and authenticated browser sessions.
Sanitization: There is no logic present in the skill to sanitize, filter, or escape the web content before it is processed by the LLM, allowing a malicious webpage to potentially hijack the agent's flow.

browser-use