stitch

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

• Third-party content exposure detected (medium risk: 0.30). In stitch-loop, the agent reads outsider-authored free text from the repo file next-prompt.md (YAML page + markdown body) each iteration and then feeds the prompt into Stitch via generate_screen_from_text (Step 3), so an outsider who can write/submit that file can inject.

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

• Potentially malicious external URL detected (high risk: 0.80). The react-components skill instructs running npm install at runtime to enable validation tools and the included package-lock.json references a tarball URL (https://registry.npmjs.org/@swc/core/-/core-1.15.8.tgz) that would be fetched and install/execute remote code required for the skill.