design-consultation

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill's Phase 2 Research explicitly instructs the agent to run WebSearch and to use the browse binary to visit public websites (e.g., the $B goto "https://example-site.com" / screenshot / snapshot flow) and to "analyze: fonts actually used, color palette…" — ingesting untrusted, public third‑party content that the agent is expected to read and that will materially influence design decisions and next actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill's SETUP step downloads and executes a remote installer via curl -fsSL "https://bun.sh/install" (saved to a temp file and run with bash), meaning remote code is fetched and executed at runtime.