file-reading

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

• Third-party content exposure detected (high risk: 0.70). This skill routes “read or summarize a file” by extension and then invokes extract scripts (e.g., org/public/core/pdf/extract.py), so at runtime the LLM ingests the contents of the user-supplied file; if that file is authored by an outsider, its body text becomes LLM-readable prose via the extract/cat path (outsider file content → LLM context).