[folathecoder]

Summary

Completes the automated and manual testing infrastructure for Aldwin 0.0.1. Every test now runs inside an autouse temporary ALDWIN_HOME and HOME with outbound AF_INET/AF_INET6 sockets blocked by default, backed by shared FakeModel, FakeHost, agent_session, and host_client fixtures, a full integration suite, a dedicated CLI rendering suite, AST architecture guards, CI coverage artifact upload, and the clean-machine manual test script in docs/manual-test.md.

Ticket

• Parent: ENG-384: 8. Testing: automated and manual
• Dev plan: ENG-401: Dev: Testing: automated and manual
• Test plan: ENG-404: Test plan: Testing: automated and manual

Closes ENG-384

Changes

• Fixtures and isolation (tests/conftest.py, tests/fakes/model.py, tests/fakes/host.py, tests/test_conftest_guards.py):
  • Made aldwin_home autouse=True (redirects ALDWIN_HOME and HOME into tmp_path, clears OPENAI_API_KEY and TAVILY_API_KEY in non-live runs, and initializes ensure_home() unless marked @pytest.mark.uninitialized_home).
  • Added no_network (autouse=True) blocking outbound AF_INET and AF_INET6 sockets while permitting AF_UNIX socket.socketpair() for asyncio and prompt_toolkit.
  • Added agent_session, host_client, fake_host, collect, and auto_approve fixtures/helpers, plus top-level FakeModel turn builders (text(..., deltas=N), tool_call, final, raise_, hang) and FakeHost.requests tracking.
• Integration suite (tests/integration/test_full_turns.py):
  • Added end-to-end tests through AldwinClient -> AldwinHost -> aldwin.core covering 5-delta streaming, run_command approval (yes and no), Esc stop during stream and during approval (< 1 s), model error handling, max_steps=15, crash after step 2 followed by reload and "continue", /new, and a 50-turn conversation verifying <= 20 user turns per model call and 1 terminal event per turn.
• CLI rendering suite and architecture guards (tests/cli/test_render_suite.py, tests/test_architecture.py):
  • Added tests/cli/test_render_suite.py covering welcome/header, dimmed history and divider, streamed markdown with code fences, exact tool cards (✓ and ✗), approval panel, end and stopped lines, all 6 ErrorCode messages, plain mode (\x1b[ and \a absent), and unknown slash command handling.
  • Upgraded tests/test_architecture.py with AST + text import verification (check_forbidden_imports), builtin_tools() 6-name verification, deliberate bad-import detection, CI matrix/artifact assertions, and docs/manual-test.md completeness checks.
• Live checks, CI, and manual script (tests/live/test_openai_live.py, .github/workflows/ci.yml, docs/manual-test.md):
  • Added @pytest.mark.live test_tavily_live_search.
  • Configured .github/workflows/ci.yml to generate and upload coverage.xml via actions/upload-artifact@v4.
  • Added docs/manual-test.md with clean-machine setup instructions for macOS and Linux, the 13-step verification script, and the platform test log template.

How it was tested

• Automated checks:
  • uv run ruff check src tests (passed)
  • uv run ruff format --check src tests (69 files formatted)
  • uv run mypy src tests (passed, 65 source files)
  • uv run pytest --cov=aldwin --cov-report=term-missing --cov-fail-under=90 (233 passed, 2 deselected in 3.71s, 92.43% total coverage; protocol/ 100%, host/ 95-96%, core/tools/ 92-100%, core/memory.py 95%)
  • OPENAI_API_KEY="" TAVILY_API_KEY="" uv run pytest -m live -v (2 skipped)

Notes for reviewers

• Clean-machine manual test runs on macOS and Linux (docs/manual-test.md section 3) will be executed and logged on ENG-384 as part of pre-release verification (ENG-385).