hive-double-checklisted

# Double Check The most dangerous review is the one you do on your own work. A model that just wrote a solution is the worst-placed reviewer of it: it shares every blind spot, every wrong assumption, and every "obviously correct" leap that produced the bug in the first place. Re-reading your own reasoning mostly re-confirms it. This skill fixes that by first getting a **second opinion from a genuinely different reasoning system** — a different AI provider's CLI agent, running locally on this machine — and then refusing to stop at the first answer. The two agents argue it out: the verifier attacks the work, the host defends what's right and fixes what's wrong, and the loop continues until both are honestly satisfied. Convergence, not exhaustion. It is **host-agnostic**. Whether this session is running in Claude Code, Codex, Gemini CLI, or Cursor, the skill detects the host and tries a *different* provider first. If every independent provider is missing or unusable, it launches a brand-new agent from the host's own provider with a clean context. That fallback is weaker than cross-provider review, but still better than asking the authoring context to review itself; always label it as a same-provider fresh-context check. **"Different provider" means a different underlying model lab, not just a different CLI.** `codex` is OpenAI, `claude` is Anthropic, `gemini` is Google. `cursor-agent` is the exception — it runs whatever model `--model` selects, so a Cursor session running So