learn-craft

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

• Third-party content exposure detected (high risk: 0.95). The skill downloads PDFs from libgen.li at runtime (Stage 2) and then reads them cover-to-cover via the Read tool (Stage 3), so the LLM ingests outsider-authored book text from a public, unvetted source.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly runs the bundled scripts/libgen.py at runtime which queries and downloads PDFs from https://libgen.li (and follows one-time get.php/CDN redirects), and those fetched books are required inputs that directly drive subagent reading and the skill's synthesized prompts/outputs.