web-design

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

• Third-party content exposure detected (medium risk: 0.65). At runtime the agent’s LLM context can be influenced by the upstream sub-skill files fetched from the outsider GitHub repo (scripts/fetch-upstream-skills.sh sparse-checkout + cp -R copies that repo’s SKILL.md/instructions/content into local files, which the agent then uses), i.e., outsider-authored free text from MengTo/Skills.

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

• Potentially malicious external URL detected (high risk: 0.90). The fetch script (scripts/fetch-upstream-skills.sh) clones https://github.com/MengTo/Skills.git at runtime and copies upstream skill folders (including SKILL.md), which the agent is instructed to treat as authoritative and therefore can directly control agent instructions/prompts.