arch-design

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md requires using web_search/fetch_content to ground research in “current web sources,” which can ingest outsider-authored free text from public web pages into the agent’s LLM context at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires loading a remote companion skill at runtime and provides the install command "npx skills add singh-gur/agent_skills --skill draw-diagram -g -y", which would fetch and execute remote package code that directly controls drawing/instruction behavior.