paw-cra-agent-designer

SUSPICIOUS. The skill is mostly coherent with a visual-design purpose and uses largely official tools and endpoints, but it handles API credentials, reads local project memory automatically, and mentions an optional low-trust loopwind dependency with pipe-to-shell installation. Overall risk is driven more by supply-chain and credential-handling exposure than by clear malicious behavior.

The fragment appears to implement legitimate HTML-to-image rendering and AI background generation, with no clear indicators of intentional malware. The main risks are raw HTML/URL interpolation, possible unintended external requests during rendering, insufficient input validation, and reduced browser isolation from --no-sandbox. Validate and contextually escape all variables, restrict URL schemes and outbound hosts, constrain filesystem paths and dimensions, and use sandboxing where operationally possible.