The Agent Skills Directory

[SAFE]: The skill incorporates comprehensive security guidelines that prioritize data privacy and environment integrity. It successfully implements protections against credential leakage and data exfiltration.
[EXTERNAL_DOWNLOADS]: Fetches configuration and project snapshots from Google's official GitHub repositories and the Agent Skills specification domain. These references are used to maintain structural consistency with established standards.
[CREDENTIALS_UNSAFE]: Prohibits the inclusion of sensitive data such as API keys, session cookies, or signed URLs in any generated reports or code repositories. It provides clear instructions for the agent to use environmental variables instead of hardcoded secrets.
[INDIRECT_PROMPT_INJECTION]: The skill manages the attack surface associated with processing external brand materials through a mandatory evidence chain:
Ingestion points: Analyzes user-provided brand assets, target audience definitions, and design constraints within SKILL.md and related references.
Boundary markers: Employs explicit [metric] placeholders to differentiate between verified data and generated design suggestions.
Capability inventory: Operations are restricted to local document generation; execution or upload tasks are deferred to specialized management skills.
Sanitization: Instructions mandate the replacement of actual customer names, phone numbers, and order IDs with demo values, and require the scrubbing of file metadata before any external transmission.

stitch-ui-style