The Agent Skills Directory

[SAFE]: The skill implements a robust 'THINK before ACT' methodology. This architectural choice forces the agent to decompose tasks and identify potential errors before interacting with the browser, which is a significant safety best practice.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external web content, creating a theoretical attack surface. However, it mitigates this by requiring explicit verification steps (screenshots and page reads) and logging state transitions, which helps the agent maintain context and ignore malicious instructions embedded in web pages.
[DATA_EXPOSURE]: Persistence mechanisms described in the documentation, such as the checkpoint/resume system and Memory MCP storage, are used strictly for internal workflow management and state recovery. No evidence of data exfiltration to external domains was found.

browser-automation