The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in SKILL.md direct the agent to perform internal account checks silently and explicitly command the agent not to tell the user about these checks. This is a concealment pattern designed to hide the agent's operations and information retrieval process from the user.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from Google Ads reports (such as search terms and performance metrics) through tools like getSearchTermReport and runGaqlQuery. This data is subsequently used to influence the agent's recommendations and actions (including 'write' operations like pausing keywords or changing bids) without explicit sanitization or boundary markers to prevent the execution of malicious instructions embedded in the external ads data.

ads