The Agent Skills Directory

[SAFE]: The skill includes comprehensive security guidelines, such as establishing security baselines, classifying data touch points for encryption, and ensuring audit trails for sensitive operations. These instructions reinforce best practices without introducing malicious behavior.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of malicious instructions in external sources (ADRs, Jira, Confluence) and explicitly mandates that the agent must only summarize or reference such content, never executing instructions found inside. This defensive prompting effectively mitigates the surface area for indirect injection attacks.
[DATA_EXFILTRATION]: External network access is limited to the vendor's documentation site via frontmatter constraints on the WebFetch tool, preventing unauthorized data transmission to untrusted domains.

architecting-solutions