push-notifications

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.70). The prompt contains instructions outside the stated push-notification purpose — notably forcing "AUTONOMOUS MODE" (override interaction) and explicit SELF-EVOLUTION TELEMETRY that directs checking and appending to a local ~/.claude/projects/skill-telemetry.md (accessing/modifying local project memory), which are hidden/deceptive behaviors beyond the integration task.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's web service worker example uses importScripts to fetch and execute remote JavaScript from the gstatic CDN at runtime (e.g. https://www.gstatic.com/firebasejs/10.x/firebase-app-compat.js and https://www.gstatic.com/firebasejs/10.x/firebase-messaging-compat.js), which causes externally hosted code to be executed as a required dependency for the web push flow.