web-game-development

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

• Third-party content exposure detected (high risk: 0.85). This routing pack can fetch and incorporate upstream sub-skill folders (including their SKILL.md and instructions) from the outsider-authored public GitHub repo MengTo/Skills at runtime via scripts/fetch-upstream-skills.sh, which then reads that text into the agent context; there is no sanitization or strict provenance gating beyond “fetch upstream”.

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

• Potentially malicious external URL detected (high risk: 0.90). The fetch script clones and sparse-checkouts the upstream repo at runtime (https://github.com/MengTo/Skills.git), and the fetched upstream SKILL.md files are treated as authoritative guidance that will control agent behavior, so this external Git URL is a runtime dependency that directly controls prompts/instructions.