when-optimizing-agent-learning-use-reasoningbank-intelligence

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

• Potentially malicious external URL detected (high risk: 0.80). I flagged the npx invocation "npx claude-flow@alpha hooks pre-task" / "npx claude-flow@alpha hooks post-task" because npx fetches and executes the remote claude-flow@alpha package at runtime and those hook commands can register/execute code that directly controls agent instructions.