human-ai-system-builder

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

• Potentially malicious external URL detected (high risk: 0.80). The skill requires running compose-agentsmd before responding and its agent-ruleset.json names the external rules source "github:metyatech/agent-rules@HEAD", which at runtime can fetch and inject remote rule content that directly controls agent instructions (github:metyatech/agent-rules@HEAD).