The Agent Skills Directory

[Secure Data Handling in Code Interpreter]: The skill instructions proactively address potential command injection risks by advising the agent to use json.dump and json.load when generating Python scripts for the sandbox. This method is preferred over string concatenation or shell heredocs, as it safely handles complex data containing special characters or unbalanced quotes.
[Sanitized Application Output]: The interactive HTML template is designed with cross-site scripting (XSS) prevention in mind. It uses safe DOM methods like textContent to render comparison claims and includes specific instructions to escape JSON payloads (e.g., replacing </ with \/<) before embedding them in script tags.
[Verified Information Workflow]: The skill implements a robust verification model that distinguishes between user-provided and web-sourced claims. By requiring an explicit human review step for external research, the skill reduces the risk of presenting unverified or inaccurate AI-generated content as settled fact.
[Least Privilege Branding Implementation]: The branding logic for HTML and PowerPoint exports is constrained to specific CSS variables and library-controlled formatting. This targeted approach allows for customization without granting the agent broad or unsafe access to presentation file structures.

competitive-battlecard-builder