iterative-dev-refactorlisted

# Refactor Scan **Opus checks, workers only fix.** Opus runs every scan and classification itself — AST walk, grep, or `wc`, whichever fits — and decides what changes. The worker NEVER checks or scans; it receives ONE concrete refactor and implements it. **One Step at a time.** Per Step: run its scan → dispatch the fix through workers (one worker per coherent unit, never bundle unrelated refactors) → evaluate the worker's plan against your own model → Go → review the diff → recap → merge. Step N's fix is merged before Step N+1 starts. Never scan ahead. **Run autonomously — report once.** No user stop between Steps; drive the run end-to-end and give ONE consolidated prose summary at the very end (per Step: what was found, what was refactored + merged). Every report is written in German; all artifacts (code, DOCS.md) stay English. **Thresholds are fixed.** The numbers below are exact — never soften one to fit a project; only the way you write the scan adapts. Cosmetic LOC shrinking (trimming blanks, merging comments) is never a split. ## Scope ASK the user which directory to refactor (the source root — `src/` or a chosen subtree). ## Phase 1 — Architectural Form ### Step 1 — Placement Is every top-level module in the right place? A `src/*.py` at the root (skip `__init__`) is justified only if ≥2 subdirectories import it OR an external entry-point loads it (`python -m x`, a uvicorn `x:app` path, `mitmproxy -s`). Imported by a single subdir with no entry-point → move it