stitch-react

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The workflow requires embedding Google Cloud Storage signed URLs (htmlCode.downloadUrl / screenshot.downloadUrl) verbatim into curl/bash commands and fetch steps, which are secret, time-limited tokens and thus forces the LLM to handle and output sensitive credentials.

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

• Third-party content exposure detected (high risk: 0.85). The runtime path described in Phase 0 fetches Stitch screen HTML via mcp__stitch__get_screen and then downloads the signed htmlCode.downloadUrl into .stitch/designs/{page-slug}.html, which is outsider-authored free text (public/third-party Stitch content) that is then read/parsed in later phases for conversion.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly downloads Stitch HTML and screenshots at runtime from Google Cloud Storage signed URLs (htmlCode.downloadUrl and screenshot.downloadUrl, fetched via scripts/fetch-stitch.sh) and then reads that HTML to drive the automated conversion to React components, so external content fetched at runtime directly controls the agent's output.