puda-manual-ui

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

• Third-party content exposure detected (medium risk: 0.65). At runtime, the required workflow captures and renders OUTSIDER text from puda process stdout/stderr (subprocess.run(..., capture_output=True, text=True) then concatenates and stores output from proc.stdout/proc.stderr), and this output is then displayed in the Streamlit UI via the command log (st.text(item["output"]...)); since the LLM agent would ingest any such displayed text, any injection can originate from whatever puda emits (untrusted command-line tool output).