scrape-analyze-page

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill explicitly reads metadata/cleaned HTML and must output extracted fields (including the source URL) verbatim into JSON and summaries, so any secrets or tokens present in meta.json or the page would be included in the agent's output.

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

• Third-party content exposure detected (high risk: 0.85). The skill reads meta.json (containing the source URL and other metadata) and then ingests cleaned HTML produced from the saved page HTML; since the saved HTML is an outsider-authored web page (public detail page content) and is converted into LLM-readable prose via clean_html.py and then parsed by extract_metadata.py, it can carry indirect prompt-injection text.