stitch::react-native

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly requires calling MCP to retrieve download URLs and then embedding those exact URLs into shell commands (bash scripts/fetch-stitch.sh), which forces the LLM to handle and output potentially sensitive presigned/download tokens verbatim.

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

• Third-party content exposure detected (high risk: 0.85). The required workflow downloads Stitch HTML/screenshots from GCS-supplied download URLs via scripts/fetch-stitch.sh (curl at runtime), and then the agent opens/reads the downloaded .stitch/designs/*.html to extract tokens (outsider-authored free text → local file content ingested into the LLM context during parsing/extraction).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly downloads external design HTML and screenshots at runtime (e.g., using bash scripts/fetch-stitch.sh with the Stitch-provided URLs "[htmlCode.downloadUrl]" and "[screenshot.downloadUrl]=w{width}"), and the fetched HTML is parsed to extract Tailwind/theme tokens that directly control generated code and agent outputs.