RUL-002 — No mocking — real infra, real implementations only

Ruling

No unittest.mock, no Mock/MagicMock/AsyncMock/patch, no hand-rolled Fake/Stub/Dummy classes, no monkeypatch, no @pytest.mark.skip/skipif. Tests run against real Redis/Neo4j/gateway/kernel. The only sanctioned test double is the LLM boundary (swap client.call_llm / inject _call_llm=fn).

Evidence

• Global CLAUDE.md engineering standard + project Hard Rules (AGENTS.md) both state this as non-negotiable; user memory user-mohamed: "strict parallel test isolation (no mocks/stubs/skips/monkeypatch — real infra only)".
• .claude/hooks/no-doubles-py.sh (PreToolUse Write/Edit guard) and .claude/hooks/forbidden-doubles-stop.sh (Stop-hook reminder) both exist specifically because the pattern kept recurring.

Enforcement

• gate scripts/test-review.py T1
• gate scripts/audit_test_purity.py monkeypatch
• codetools-hook no-doubles-py

Verification

scripts/test-review.py rule T1 ("No mocking") flags unittest.mock imports and Mock/MagicMock/AsyncMock/patch/mocker usage; scripts/audit_test_purity.py scans for the monkeypatch pattern class (plus Fake/Stub/skip) and exits non-zero on any hit; codetools hook no-doubles-py (wired in claude/project-hooks.json; formerly .claude/hooks/no-doubles-py.sh) refuses a Write/Edit that introduces a double.