ralph

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches GitHub issue bodies and sub-issues via the gh CLI in ralph-once.sh and afk-ralph.sh (building prd_body and sub_issues) and injects those untrusted, user-generated GitHub issue contents into the Claude/docker prompts for autonomous implementation, so third-party issue text can directly influence agent actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's scripts call the GitHub CLI (gh) at runtime to fetch PRD and sub-issue bodies and then inject that fetched content into the agent prompt (i.e., the GitHub Issues API endpoints such as https://api.github.com/repos///issues/ are used to supply instructions), so remote GitHub issue content directly controls the model's prompts and is a required runtime dependency.