fix-looplisted

# Fix Loop — Parallel Check → Fix → Repeat Run checks across all failure dimensions simultaneously using minimal-context agents, fix every failure found, then re-verify. Loop until everything passes or `$1` iterations are exhausted (default: 5). **Max iterations**: `${1:-5}` --- ## Quick Start with `/goal` (Recommended) Use `/goal` to run this loop hands-free — no per-turn prompting required: ``` /goal all build, test, lint, and type checks pass and the exit code is 0, or stop after ${1:-5} iterations ``` Setting that goal executes Steps 0–4 automatically, with the evaluator (Haiku) checking the condition after each turn. When all checks pass, the goal clears. When stuck or at max iterations, it stops and reports. The manual loop in Steps 1–5 below is the fallback when you need step-level control or want to inspect intermediate state between iterations. --- ## When to Use - After implementing a feature — run until green before opening a PR - When `quality:does-it-work` reports failures you want auto-resolved - When you have a large test suite with scattered failures across modules - Faster than running fixes manually: parallel agents, minimal context per agent, automatic re-verification **Not a replacement for** `code:is-it-ready` — that does architectural review. This does build/test/lint repair. --- ## Step 0 — Project Inventory Before launching agents, establish ground truth: ```bash # 1. Identify project type(s) ls -1 pyproject.toml package.json build.grad