Summary

Cluster H-1 of #643 — Windows console default codepage (cp1252/cp437) cannot encode the box-drawing and em-dash glyphs the wizard prints, so mm init crashes at the very first banner line on a fresh Windows runner.

This is a production bug, not a test-only artifact: every Windows user hits it on the first mm init invocation.

Stack from the failing CI run (sha 88c3e81):

init_cmd.py:2781  click.secho("  ─────────────", fg="cyan")
click/utils.py:321  file.write(out)
UnicodeEncodeError: 'charmap' codec can't encode character '─' (U+2500)

Approach

Spot-fixing line 2781 leaves every other unicode glyph (em-dash, future additions) as a latent crash. Reconfigure sys.stdout / sys.stderr at the CLI entry point (the cli() group callback runs before any subcommand emits output), gated on sys.platform == "win32" so POSIX is a no-op:

if sys.platform == "win32":
    for stream in (sys.stdout, sys.stderr):
        try:
            stream.reconfigure(encoding="utf-8", errors="replace")
        except (AttributeError, io.UnsupportedOperation):
            pass

errors="replace" degrades a missing glyph to ? instead of crashing — strictly better than the current behavior.

Test plan

• POSIX unchanged: uv run pytest -m "not ollama" packages/memtomem/tests/test_cli.py packages/memtomem/tests/test_cli_status.py packages/memtomem/tests/test_init_cmd.py — 355 passed.
• uv run pytest -m "not ollama" packages/memtomem/tests/test_cli_index_noop_e2e.py — 2 passed.
• uv run ruff check + uv run ruff format --check on the touched file.
• CI test (windows-latest) will turn test_init_index_search_via_subprocess from FAIL → PASS (cluster H-1 of the Windows compat sweep).

Cluster H-2 (test_force_overrides_db_lock) is a separate root cause and ships in a follow-up PR.

🤖 Generated with Claude Code