A

Final Refactor Audit

Baseline: f2eb2c7a1ac56c913d4a44ab24d5c3a2b553007c

Branch: codex/macro-refactor-from-f2eb2c7

Result#

This pass stops after five behavior-preserving refactor commits. The branch keeps the existing product surface and improves the parts with clear boundary pressure:

  • setup rendering no longer hides a terminal toolkit in one command renderer;
  • app composition reads as grouped manual dependency wiring;
  • sync execution no longer repeats ledger write mechanics across foreground and background paths;
  • raw parser and validator boundaries no longer use production Any;
  • Codex app-server notification dispatch no longer owns event-construction details.

Verification#

Every committed code batch passed:

  • uv run pytest
  • uv run ruff check .
  • git diff --check

The final code batch passed 402 tests.

Remaining Large Files#

The largest remaining classes and files are not automatic refactor targets. They are mostly documented facades, orchestration services, or provider adapters:

  • services/runs/store.py
  • services/topics/mutations.py
  • services/viewer/service.py
  • workflows/page_run/service.py
  • workflows/sync/evaluation.py
  • integrations/harnesses/codex/app_server.py

Further splitting these now would be mostly taste unless a new feature exposes a specific reason-to-change problem.

Product Questions Preserved#

Behavior that may be product slop remains recorded in product-slop.md. No behavior was changed to simplify architecture.

Stop Rationale#

The branch has reached the point where the next obvious moves are smaller local preferences, not macro architecture improvements. The codebase now has stronger boundaries around the areas that were actively hard to scan or extend, and the remaining complexity appears tied to actual product behavior.