motion-graphics

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The codebase includes explicit paths that send user-provided image/data to third‑party APIs (the locate "auto" flow using GEMINI_API_KEY) and multiple runtime network fetches (tile/CDN world-atlas) plus a headless Chrome launch with --no-sandbox — these are clear data‑exfiltration / security‑weakening patterns that could leak sensitive inputs when environment keys or external sources are enabled.

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

• Third-party content exposure detected (medium risk: 0.30). The runtime path /hyperframes → phases/source/guide.md (Step 2) → phases/source/resolve.mjs is triggered when shot-plan.json.asset_needs is non-empty (news/web/tweet/image), and this sourcing uses outsider-authored free text from those sources (e.g., article text, webpage content, tweet text) to generate the shot.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill launches bake-basemap.mjs and in-page captures that (at runtime) fetch and execute remote JS/CSS and data (maplibre, topojson, world-atlas, Esri/CARTO tiles, GSAP) and also contains an optional runtime call to the Generative Language API (https://generativelanguage.googleapis.com/...) — these URLs are fetched during runtime and execute remote code or model inference used by the bake/localization flows.