The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes functionality to ingest and summarize user feedback from surveys via the surveys-responses-list tool as described in references/qualitative-feedback.md. While the skill body instructs the agent to treat this text as untrusted data, it remains a potential vector for malicious actors to provide instructions designed to influence the agent's analytical conclusions.
[COMMAND_EXECUTION]: The skill dynamically generates HogQL queries for execution via the execute-sql tool based on templates in multiple reference files. Although the instructions direct the agent to resolve identifiers through the experiment-get tool first, there is a risk of SQL injection if user-supplied strings such as feature flag keys or event names are directly interpolated into these query templates without adequate sanitization.

diagnosing-experiment-results