save

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

• Third-party content exposure detected (high risk: 1.00). The skill directly fetches and scrapes arbitrary public URLs (see scripts/extract-web.sh and scripts/extract-defuddle.sh which call Defuddle/Firecrawl on provided URLs and write to $SAVE_CTX/extracted.txt), and that untrusted web/X/YouTube content is read and used by the workflow to translate, restructure, generate metadata, and decide follow-up actions—so third-party content can materially influence agent behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill runs extract-defuddle.sh (calling defuddle parse "$URL") and extract-web.sh (falling back to firecrawl scrape "$URL") at runtime to fetch arbitrary https://... pages (the user-supplied ) and writes that fetched content to $SAVE_CTX/extracted.txt which is then read and sent to translation/restructure/review/model subagents, so remote content is directly injected into and can control agent prompts.