comfyui-gateway

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

• Third-party content exposure detected (medium risk: 0.30). The gateway ingests outsider-authored free text directly from API requests—clients can POST arbitrary inputs (e.g., prompt/negative_prompt) to /jobs, which the worker renders into ComfyUI workflow templates at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The Supabase Edge Function imports remote code at runtime from https://deno.land/std@0.177.0/http/server.ts which causes Deno to fetch and execute external module code as a required dependency, so this URL is a runtime-executed remote dependency.

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I flagged the AWS-style S3 credentials because they are literal, high-entropy keys in the document and match the formats used for real AWS access keys and secret access keys (AKIA... and the long base64-like secret). These appear directly in the production environment example and could be mistaken for usable credentials.

I ignored other credential-looking values because they are clearly placeholders or low-entropy defaults used for examples:

• minioadmin / minio default passwords (low entropy, common setup defaults).
• Strings like your-api-key, sk-admin-key, change-this-to-a-random-secret, docker-jwt-secret-change-me, webhook-hmac-secret-at-least-32-chars, strong_password, redis-password — these are examples/placeholders or simple setup passwords and not high-entropy secrets.