base-ui-migrator

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow can ingest outsider-authored free text from the target codebase itself: scripts/scan-candidates.sh <path> reads arbitrary files under the user-supplied <path> (e.g., React source/JSX comments/strings) and then the agent uses that extracted prose/strings in LLM context for triage and migration decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly invokes refresh/fetch scripts (scripts/refresh-catalog.sh and scripts/fetch-component-doc.sh) that use curl to retrieve the catalog and per-component docs from base-ui.com (e.g. https://base-ui.com/llms.txt and component docs on base-ui.com) at runtime, and those fetched documents are used to drive the agent's migration instructions — a remote URL controlling prompts.