automate

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow passes prompt/diff text to first-party CLIs (Claude Code claude -p "$(cat ...)", Codex codex exec ... with piped stdin, Gemini gemini -p ... $(cat ...), Copilot copilot -p ...) and can incorporate outsider-authored content that the script reads (e.g., PR diffs in the cookbook or untrusted files/logs), meaning the LLM ingests free text supplied via those runtime inputs.