Context

PR #465 introduced tests/test_batch_add_tag_isolation.py which redefines the _isolate_memtomem_env + _StubCtx + integration_components fixture trio that already exists almost verbatim in tests/test_multi_agent_integration.py:56–127.

Both fixtures share the same intent: real BM25-only component stack on a tmp DB, hermetic against ~/.memtomem/config.json and MEMTOMEM_* env vars. Two callers is tolerable; a third would push it over the threshold for extraction.

Scope

Move into tests/conftest.py:

• _MEMTOMEM_ENV_VARS constant
• _isolate_memtomem_env(monkeypatch) helper
• _StubCtx class
• integration_components fixture (rename to make its scope clear, e.g. bm25_only_components)

Then drop the local copies from both files. No behavior change.

Why not in #465

PR #465 was a focused 24-line removal + 170-line test addition for the broadcast-tag bug. Pulling fixture extraction into the same PR would have been scope creep (feedback_one_change_per_pr.md). Tracked here so the next caller doesn't grow a third copy.

Reference

• Reviewer comment on fix(mem_batch_add): drop post-index tag broadcast; per-entry tags isolated #465 (N1): "3번째 사용처 등장 전 conftest 로 추출 권장"
• feedback_refactor_axis.md: "중복이 semantic coupling 이면 추출"