Imported from ChiR24/Unreal_mcp (tests/unit/mcp-primitives/AGENTS.md). Install upstream with npx skills add ChiR24/Unreal_mcp --skill mcp-primitives. Copyright stays with the author.

MCP PRIMITIVES NATIVE-PARITY TESTS

Vitest tests that gate byte-for-byte parity between the TypeScript MCP primitives (src/server/mcp-primitives/) and their native mirror (plugins/McpAutomationBridge/.../Private/MCP/Primitives/). A drift here fails CI.

SCOPE

27 .test.ts files in tests/unit/mcp-primitives/ (5 named *-parity.test.ts). No subdirectories. Each file covers one primitive family and uses a GREEN/RED guard pattern (see below).

WHERE TO LOOK

CONVENTIONS

• GREEN/RED guard pattern: each test asserts the TS output matches the native fixture oracle (GREEN), then asserts a deliberately-mutated TS output does not match (RED). The RED guard catches a test that passes trivially because both sides are empty or wrong.
• The native oracle is a frozen fixture (JSON or typed literal) checked into the test file — it does not spawn the native transport. Update the fixture only when the intended native contract changes, and update both sides together.
• Parity is on shape (keys, types, enum values), not on runtime-produced values like timestamps or generated IDs.
• perActionSchemas is always false on both surfaces — assert it explicitly; a parity test that omits this check can miss a regression to per-action schema dumping.

ANTI-PATTERNS

• Do not make a parity test pass by deleting the native side; the RED guard will catch the missing oracle.
• Do not compare runtime values (ports, session IDs, timestamps) — they differ by transport and make the test flaky.
• Do not add a new primitive to one surface without adding the parity test for it in the same PR; CI will not catch the drift otherwise.
• Do not relax the GREEN/RED pattern to GREEN-only — a GREEN-only test passes when both sides are broken identically.