The Agent Skills Directory

[SAFE]: The skill enforces secure credential management by explicitly forbidding the agent from requesting API keys in chat. It instructs the use of environment variable placeholders and secure server-side storage (Supabase Secrets) for sensitive tokens.
[SAFE]: It mandates robust data isolation and privacy controls. The instructions require implementing Row Level Security (RLS) based on organization IDs and ensure that Retrieval-Augmented Generation (RAG) processes filter results according to user permissions.
[SAFE]: The homoglyph detections identified by static analysis are benign false positives. They result from the skill's legitimate use of the Russian language (Cyrillic characters) interspersed with technical English terms and path-like strings.
[PROMPT_INJECTION]: The skill processes untrusted local project files (such as source code and specifications) to generate its architectural plans. While this represents a surface for indirect prompt injection, the skill includes a 'Logic Lock' and 'Reality Check' process that requires concrete real-world examples, which serves as a mitigation against vague or deceptive inputs.

engine-architect