--- name: the-loop description: Use when you have a draft, a piece of work, or an output that needs to be genuinely *good* — not just done — and you want it refined through real cycles instead of stopping at the first attempt. Reach for it when quality matters more than speed, when first drafts aren't enough, when "make this actually good" is the job. Works on writing, plans, designs, code, decisions — anything with a draft and a standard it should meet. --- # The Loop ## Overview A skill for one job: **make the work genuinely good by iterating it against an explicit standard — then stop.** You are the craftsman who won't ship slop, and also won't polish forever. The discipline is the bar: define what "good" means, then cycle until the work clears it. **The core move:** *set the bar first, then draft → critique against the bar → revise → repeat until it clears.* The order matters. A bar before the draft turns vague "make it better" into a checklist you can actually hit. **Core principle:** *Iteration only improves work when there's something to improve against.* Looping with no standard is wheel-spinning — it burns effort and drifts. Looping against an explicit bar compounds. The bar is the whole skill; the loop is just how you reach it. **Announce at start:** "I'm running this in a loop against an explicit quality bar. First I'll define what 'good' looks like, then I'll iterate until the work hits it — not forever, just until it clears." ## When to use - A draft that needs to become genuinely good — writing, a plan, a pitch, a design, code. - High-stakes output where first-draft quality isn't enough. - "Make this actually good", "tighten this up", "this isn't there yet." - Anywhere you'd otherwise accept the first plausible version. ## When to skip - The first draft is already good enough for a low-stakes need — don't gold-plate. - You don't yet know what you're making (use moonshot to generate, or consultant to structure) — you can't iterate toward a bar you can't define. - There's no meaningful standard to hit — then there's nothing to loop against. ## The method The engine is **draft → critique → revise**, the same loop behind red-green-refactor in software and the "Reflexion" self-critique pattern in AI. What makes it work is a real bar and a *harsh* critic. ### 1. Define the quality bar — FIRST, and concretely Before drafting (or before the first revision), write down **what "good" looks like** as explicit, checkable criteria. A rubric, a checklist, 3–6 concrete tests. *"Clear to a smart 12-year-old", "every claim backed", "under 200 words", "no jargon", "a stranger could act on it without asking a question."* **This step is the whole game** — vague bars produce vague work. ### 2. Produce a draft Get a real first version down. It doesn't need to be good — that's what the loop is for. ### 3. Critique against the bar — harshly Go criterion by criterion: **pass / fail**, and for each fail, *exactly what's missing.* Be the tough critic, not the cheerleader — a soft self-critique produces no improvement (this is the documented failure of weak reflection loops). Name the gaps plainly. ### 4. Revise the failures — don't start over Fix the specific things that failed. **Don't rewrite from scratch** — that throws away what already cleared the bar and resets progress. Targeted revision compounds; full rewrites churn. ### 5. Repeat 3–4 until it clears Loop the critique and revision. Each pass, the work should measurably close on the bar. ## The loop — the stopping rule This skill *is* the loop, so the stopping rule is the point: - **Stop when the work clears the bar** — every criterion passes. That's done. Don't keep polishing past it (over-iteration is its own waste). - **Or stop at the round cap** (default **3–5**; raise it only for genuinely high-stakes work you've been told to take time on). - **Never stop *before* the bar** just because a draft "feels finished" — that feeling is usually round-one fatigue, not quality. - If a criterion *won't* clear after honest effort, say so explicitly — name what's unmet and why — rather than silently lowering the bar. The cap is what stops it burning tokens forever; the bar is what stops it quitting too early. Both are required. ## Output shape > **The bar** — the concrete criteria "good" had to meet (stated up front). > 1. **The final work** — the version that cleared the bar. > 2. **What changed** — the meaningful improvements across rounds (so the lift is visible). > 3. **Anything unmet** — any criterion that didn't clear, named honestly, with why. ## Honesty rules (non-negotiable) - **No bar, no loop.** "Make it better" with no defined standard is drift. If you can't write the criteria, stop and define them — that *is* the work. - **The critique must bite.** A self-critique that rubber-stamps the draft improves nothing. Each round must find real, specific gaps — or honestly conclude the bar is met. - **Don't move the goalposts.** Lowering the bar to declare victory is the cardinal sin. If it can't clear, say it can't clear. - **Stop at the bar.** Iterating a passing draft "just in case" wastes effort and often makes it worse. Done is done. ## Common mistakes / red flags — STOP - **Implicit bar.** Looping without writing the criteria down → it drifts and never converges. Define the bar explicitly, first. - **Soft critique.** "This is pretty good, maybe tighten a bit" → no real revision happens. Be specific and harsh: which criterion fails, and exactly how. - **Rewriting from scratch each round.** Throws away progress, resets the clock. Revise the *failures*, keep the passes. - **Running past done.** Polishing a draft that already cleared the bar. Stop. - **Goalpost-moving.** Quietly redefining "good" down to end the loop. Hold the bar; report what's unmet.