nordisk/nornir

test_history MCP tool + release test-drive gate

Plan #5. Two additions, both built additively (no proto change) so they merge cleanly alongside concurrent proto/viz work:

1. a test_history MCP tool — read a repo's per-run test summaries; and
2. a release test-DRIVE gate — a release actually runs the repo's test matrix and fails on red / regression.

Companion of bench_history (the exact template) and the existing test_results / build-free test_inventory machinery (.nornir/test-matrix.md, .nornir/test-matrix-always-visible.md).

1. test_history MCP tool

mcp__nornir__test_history — src/bin/nornir-mcp.rs : NornirMcp::test_history.

Input: { repo : string, limit? : usize }. Output: a JSON array of per-run summaries, newest-first — the SAME per-run rollup nornir test history <repo> prints, one object per run:

[ { "run_id", "repo", "ts_micros",
    "passed", "failed", "ignored", "stalled", "skipped", "listed",
    "total", "green" }, … ]

green and total are folded in by the canonical serializer so an agent never recomputes them.

Two modes (embedded XOR client), mirroring bench_history

• Embedded (NORNIR_SERVER unset): open the warehouse read-only (lock-tolerant, like nornir test history), query_test_results scoped by Repo(repo) — falling back to Run(repo) when no repo matches (same selector the CLI accepts) — then summarize_runs + cap to limit.
• Client (NORNIR_SERVER set): the server owns the warehouse lock, so the tool pulls every test_results row over the existing Viz.TestResults RPC and applies the SAME pure scope+summarize (test_history_summaries). No new RPC — reusing Viz.TestResults keeps the change additive.

The scope+summarize+limit logic is pure and shared (test_history_summaries + summarize_runs + runs_to_json), so the two modes return byte-identical JSON (proven in tests).

Canonical serialization

nornir_testmatrix::model::runs_to_json (re-exported as nornir::warehouse::test_results::runs_to_json) renders a &[RunSummary] newest-first with the derived green/total flags. RunSummary gained Serialize/Deserialize for the RPC round-trip.

2. Release test-DRIVE gate (plan #5b)

A release must EXERCISE + check its tests, not just read old results.

• release::gate::tests_gate(rows) — the pure verdict over already-collected test_results rows. Fails if there is no run, the newest run is RED (any fail/stalled), or a test that was green in the last green run is now red (a regression, named explicitly). Mirrors coverage_gate.
• release::gate::run_tests_gate(repo_root, history, aspects) — the heavy drive: it runs the matrix (test_matrix::run_full_matrix, the same engine Ops.RunTestMatrix uses), folds the fresh rows with the repo's persisted history, applies tests_gate, and returns the fresh rows so the caller persists them to the warehouse.

Wired into release_gate_all (the MCP embedded path + the server Ops.GateAll handler) behind the per-repo [gates].tests = true toggle. HEAVY — it compiles

• runs the suite — so it is invoked release/on-request only (the gate call), never on workspace load, consistent with the heavy-test policy. The freshly-run rows are persisted (append_test_results_blocking) so the 🧪 Test pane and test_history see the release run.

Tests (inject-assert)

• nornir-testmatrix model: runs_to_json newest-first + green/total folding (model unit tests).
• src/release/gate.rs::tests_gate_passes_green_fails_on_red_and_names_regression — green→Ok, no-run→red, red→Err naming the case, regression→Err naming the regressed test, stalled→red.
• tests/release_gates.rs::tests_gate_green_pass_red_fail_regression_through_public_api (fast, public API) + run_tests_gate_drives_matrix_and_passes_green_fails_red (#[ignore] heavy: scaffolds a passing + a failing cargo crate, drives the matrix end-to-end, asserts green passes / red fails).
• tests/support/mcp_harness.rs seeds two holger test runs (older green, newer red) and exercises test_history, asserting the newest (red) run leads and carries green:false — covered by mcp_every_tool_exercised_and_asserted (coverage gate goes RED if the tool is ever unexercised).

CLI ⇄ MCP ⇄ viz parity

• CLI: nornir test history <repo> (+ --json) — pre-existing.
• MCP: test_history — this addition.
• viz: 🧪 Test tab (summarize_runs/RunSummary) — pre-existing; the new gate's freshly-run rows flow into it via the warehouse.