uploading-files

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

• Third-party content exposure detected (high risk: 0.85). The skill’s fetch step downloads user-uploaded files from the outsider-controlled GitHub web upload branch (raw.githubusercontent.com/.../{branch}/{rel}) and writes them into ./.uploads/, which the agent then treats as local input—so any text content in those files becomes LLM-readable context (indirect prompt injection risk).