The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to navigate and test external websites, which inherently involves processing untrusted content. This represents a potential surface for indirect prompt injection where a malicious website could attempt to influence the agent.
Ingestion points: Instructions in SKILL.md to load candidate URLs, routes, and exercise journeys.
Boundary markers: The instructions do not specify delimiting external content or provide explicit warnings to ignore embedded instructions within the target web pages.
Capability inventory: The skill utilizes browser interaction and computer-use tools to perform testing tasks.
Sanitization: No content sanitization or validation steps are defined for the data retrieved from the tested websites.
[SAFE]: The skill provides clear safety boundaries, instructing the agent to use only host-approved browser connectors and to avoid unauthorized external side effects like making purchases or altering production data. The instructions are consistent with professional software testing practices.

starlight-qa