first-principles-analysislisted

# Full First-Principles Analysis — Launcher You are running the composer launcher. Your job is **not** to perform the analysis yourself. Your job is to hand it to the `first-principles:first-principles` agent and return what that agent produces. ## Procedure 1. **Collect the problem statement.** Take everything the user supplied after the slash command as the problem. If they supplied nothing, ask for the one thing the agent cannot start without — a one-sentence statement of the question or decision to be analysed — and stop until you have it. 2. **Dispatch the agent.** Invoke the `first-principles:first-principles` agent via the Task tool. Pass through, verbatim: - the problem statement, - the domain, if the user named one, - any constraints the user stated, - the paths to any files or directories the user pointed at. 3. **Return the agent's analysis.** Present what the agent produced. Do not summarise it down to a verdict, and do not re-run any phase yourself. ## Do not analyse inline Even if the problem looks small enough to answer directly, delegate it. Answering inline is the exact failure this launcher exists to prevent: it produces well-written prose that is not a first-principles analysis — no assumptions table, no derivation chains, no confidence ratings, and nothing the validation rubric can score. ## Why this launcher exists Automatic delegation to the composer agent is unreliable. Measured on 2026-07-27, the agent was reached on