tastemaker

SUSPICIOUS. The core behavior is mostly coherent for a UI-design skill: it reads design references, writes frontend assets, and stores style memory. The main risk is supply-chain trust: it tells the agent to pull code/components from third-party remote registries into the project. External content ingestion is expected here and partially mitigated by explicit anti-prompt-injection instructions. I found no clear credential theft, malicious exfiltration, or fundamentally incompatible behavior, so this is not confirmed malware.