design-import

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

• Third-party content exposure detected (high risk: 1.00). The Phase 1 orchestrator explicitly fetches a claude.ai/design handoff bundle (via the provided URL), extracts and reads README.md and chats/*.md plus the primary HTML prototype from that tarball—untrusted, user-originated content—which is parsed into the normalized payload and then used to drive token reconciliation and scaffold/adapt code, so third-party content can materially influence agent actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches and extracts a Claude Design handoff bundle from URLs like https://claude.ai/design/abc123 at runtime and explicitly reads README and chats/*.md and HTML prototypes which are parsed and injected into agent prompts—meaning remote bundle content can directly control agent instructions.