paper-glossary

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested via the runtime paper-content boundary: plan_glossary.py loads *_raw_sections.jsonl (from --source-manifest / --raw-sections) and extracts record["text"] into effective_prose_text(...), which is then used to generate the LLM-facing candidate pool and shortlist.