session-titles

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

• Potentially malicious external URL detected (high risk: 0.80). The skill uses the Anthropic API at runtime via the @anthropic-ai/sdk (e.g. https://api.anthropic.com) — scripts like evolve-prompt.ts and eval-quality.ts send context to Sonnet/Haiku models and ingest their outputs to generate or mutate systemPrompt/instructions, so remote responses directly control prompts.