iteratelisted

# Iterate — Convergence Loop Autonomous brainstorm→fix→re-scan cycle. Runs until the codebase converges (no more findings) or hits the round limit. ## How It Works ``` iterate |-- Round 1: brainstorm (5 parallel scans) → apply → auto (fix all) |-- Round 2: brainstorm (re-scan) → apply → auto (fix remaining) |-- Round 3: brainstorm (re-scan) → 0-2 findings → done | +-- Converged: "Codebase clean after N rounds." ``` Typical convergence: 3-4 rounds. Each round finds fewer issues as the codebase improves. ## Entry 1. Set auto-active flag (same as auto mode) 2. Check pre-flight: build passes, prd.json archived if needed 3. Start Round 1 ## Round Execution Each round follows this sequence: ### Step 1: Brainstorm (scan) Run the full brainstorm Phase 1 scan (5 parallel agents: dead code, errors, complexity, deps, live QA). ### Step 2: Evaluate findings Count findings by priority: - If **0 findings** → converged, exit loop - If **1-2 low-priority** findings → fix inline without creating stories, then exit - If **3+ findings** → continue to Step 3 ### Step 3: Apply (create stories) Write findings to prd.json as stories (brainstorm apply). Deduplicate against existing stories. ### Step 4: Auto (execute) Work through all pending stories using the auto execution flow: - Context loading → implement → typecheck → build → visual verification → mark complete - Commit every 3 tasks ### Step 5: Check round limit - If rounds < max (default 4) → start next round - If