Recommendations for cleanup

1. Kill the test bloat and mock avalanche

• Problem: tests are ~1.8x the size of the source, and 336/417 test files import unittest.mock. Many test files are clearly "coverage" padding (e.g. test_full_coverage.py, testcoverage.py,
  test*_extended.py, test_remaining_coverage.py).
• Cleanup: Merge coverage-only tests into real behavior tests. Delete files whose only purpose is to hit a line-count target. Replace MagicMock/patch heavy files with integration tests against the actual
  plugin/module behavior.

2. Stop print()ing from production code

• Problem: 34 source files use print(). Some are demos/CLI helpers, but print() also appears in terminal_utils.py, cli_runner.py, plugin menus, and ask_user_question renderers.
• Cleanup: Route all output through the message bus / Rich renderer / logging. Remove or gate demo print() calls behind main blocks.

3. Replace silent except: pass with structured logging

• Problem: 39 source files have bare pass in exception handlers. This makes failures invisible.
• Cleanup: Every pass in an except block should log at debug or warning level, or re-raise if it is truly unrecoverable. Add a lint rule for bare except: pass.

4. Remove or finish TODOs

• Problem: only 4 TODOs, but they sit in http_utils.py, hook_engine/aliases.py (3x), and one is a docstring example.
• Cleanup: Resolve the hook engine alias TODOs or delete them if no longer needed. Remove the http_utils.py TODO by either implementing RetryingClient(httpx.Client) or deleting the comment.

5. Consolidate duplicated config/setting logic

• Problem: config.py is 2209 lines and contains many near-identical getter/setter pairs (temperature, top_p, seed, banner colors, output level, etc.). command_line/set_menu_catalog.py, model_settings_menu.py,
  set_menu_values.py, etc. duplicate catalog logic.
• Cleanup: Generate config keys/getters/setters from a single schema dataclass or dict. The existing set_menu_schema.py is a start but is not used to drive the actual config functions.

6. Unify HTTP clients

• Problem: httpx and requests are both used; requests appears in 8 source files (mostly OAuth plugins) and httpx in ~25. http_utils.py has a RetryingAsyncClient but no sync counterpart, and claude_cache_client.py
  builds its own client.
• Cleanup: Provide one async and one sync client factory in http_utils.py and make all plugins use them. Drop requests entirely if possible, or isolate it to one adapter.

7. Refactor the plugin registration explosion

• Problem: many plugins register only custom_command + custom_command_help and are essentially menu wrappers. Others register model types or agents. The surface area is huge.
• Cleanup: Introduce a Plugin base class or decorator that declares which phases it uses. Move boilerplate help/command registration into a shared mixin. This would shrink many register_callbacks.py files.

8. Split command_line/ into UI and command logic

• Problem: command_line/ has 62 files mixing Textual screens, completion, command handlers, and form wizards.
• Cleanup: Separate screens/ (Textual UI), commands/ (slash command implementations), and completions/ (prompt-toolkit completers). The current flat layout makes navigation hard.

9. Remove pickle session storage or sign it properly

• Problem: session_storage.py stores message history as pickles with a note that the legacy signed format was removed. Pickles are unsafe and a known footgun.
• Cleanup: Either sign pickles with HMAC again, or better, serialize to JSON/msgpack. pydantic-ai messages should be serializable to JSON already.

10. Reduce monkey-patching in pydantic_patches.py

• Problem: 543 lines of runtime monkey patches for pydantic-ai, termflow, prompt_toolkit, etc.
• Cleanup: Push upstream-compatible changes into wrapper classes (e.g. custom ToolManager, custom provider clients) instead of patching internals. Keep only patches that have no other hook.

11. Tighten environment variable access

• Problem: os.environ / os.getenv is scattered across 40+ files with inconsistent truthy checks.
• Cleanup: Centralize env access in config.py or provider_credentials.py. Provide typed helpers like get_bool_env, get_str_env.

12. Revisit the TUI/Textual dependency

• Problem: the repo is heavily investing in a Textual TUI (feat(tui) commits are everywhere in June) but pyproject.toml does not declare textual as a dependency.
• Cleanup: Either add textual (and any other TUI deps) to pyproject.toml or make the TUI code lazy/degradable when Textual is absent.

13. Simplify the model factory provider matrix

• Problem: model_factory.py has provider-specific code for ~10 providers with duplicated API-key/endpoint resolution logic.
• Cleanup: Drive provider construction from a declarative table (type → provider class, env var, default endpoint) and a single resolver function.

This is a project-wide cleanup list rather than one scoped task, with relevant entry points including config.py, http_utils.py, session_storage.py, pydantic_patches.py, pyproject.toml, and model_factory.py. First choose one numbered section and inspect its named files; no single test or completion criterion is provided, so the selected refactor needs its own scope and validation plan.