Cavendish

Query ChatGPT Pro and Deep Research from the shell via the Cavendish CLI. Run prompts, attach files (including a repo tar.gz + git diff for an architecture-level code review with full repo context), and wait for the result through detached jobs. EXPLICIT TRIGGER: "ask ChatGPT", "ask GPT", "run Deep Research", "get GPT's opinion", "send the repo to ChatGPT", or equivalent wording. DEFAULT — `ask` is the right default for almost any non-trivial question where ChatGPT Pro's perspective would add value: design decisions, architecture-level code review with full repo context, low-confidence technical claims, second-opinion checks, or whenever a second LLM view is useful. Attach tar.gz / diff / individual files as the question warrants. USE deep-research INSTEAD when the task is a broad, multi-source investigation — comparing 3+ vendors / technologies with possibly stale training data, gathering current market / pricing / regulatory data, or whenever the user asks for a thorough or comprehensive investigation acros