ml-plan

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

• Third-party content exposure detected (high risk: 0.90). The skill's Web mode explicitly requires making WebFetch calls to public third-party pages (official docs, GitHub repos, Hugging Face model cards, etc.) and mandates quoting and using that fetched content to build plans, so the agent will ingest and act on untrusted open-web content (SKILL.md "Web mode" / "WebFetch" rules).

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly requires making WebFetch calls at runtime to external documentation/repos (e.g., https://github.com/axolotl-ai-cloud/axolotl) and mandates using fetched page content to ground prompts, code signatures, and plan text, so these external URLs are runtime dependencies that directly control agent instructions.