harness-codebase-reverse-spec

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

• Third-party content exposure detected (high risk: 0.75). The required workflow ingests the user-selected target codebase contents (e.g., files and tests from the repo being reversed) into the agent’s LLM context; because those sources are outside the operating user’s authored text, they qualify as outsider-authored free text (code/comments/README/test descriptions) via the repo-reading/extraction phases.