test(e2e): retire test-strict-tool-call-probe.sh

.github/workflows/regression-e2e.yaml

@@ -21,7 +21,7 @@ on:
       jobs:
         description: >-
           Comma-separated regression job names to run (empty = all).
-          Valid: dashboard-remote-bind-e2e,gateway-health-honest-e2e,gateway-drift-preflight-e2e,openshell-version-pin-e2e,onboard-inference-smoke-e2e,strict-tool-call-probe-e2e,model-router-provider-routed-inference-e2e,openclaw-plugin-runtime-exdev-e2e,whatsapp-qr-compact-e2e
+          Valid: dashboard-remote-bind-e2e,gateway-health-honest-e2e,gateway-drift-preflight-e2e,openshell-version-pin-e2e,onboard-inference-smoke-e2e,model-router-provider-routed-inference-e2e,openclaw-plugin-runtime-exdev-e2e,whatsapp-qr-compact-e2e
         required: false
         type: string
         default: ""
@@ -49,7 +49,6 @@ jobs:
       gateway_drift_preflight: ${{ steps.select.outputs.gateway_drift_preflight }}
       openshell_version_pin: ${{ steps.select.outputs.openshell_version_pin }}
       onboard_inference_smoke: ${{ steps.select.outputs.onboard_inference_smoke }}
-      strict_tool_call_probe: ${{ steps.select.outputs.strict_tool_call_probe }}
       model_router_provider_routed_inference: ${{ steps.select.outputs.model_router_provider_routed_inference }}
       openclaw_plugin_runtime_exdev: ${{ steps.select.outputs.openclaw_plugin_runtime_exdev }}
       whatsapp_qr_compact: ${{ steps.select.outputs.whatsapp_qr_compact }}
@@ -98,12 +97,6 @@ jobs:
             echo "onboard_inference_smoke=false" >> "$GITHUB_OUTPUT"
           fi
 
-          if [ -z "$normalized" ] || includes_job "strict-tool-call-probe-e2e"; then
-            echo "strict_tool_call_probe=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "strict_tool_call_probe=false" >> "$GITHUB_OUTPUT"
-          fi
-
           if [ -z "$normalized" ] || includes_job "model-router-provider-routed-inference-e2e"; then
             echo "model_router_provider_routed_inference=true" >> "$GITHUB_OUTPUT"
           else
@@ -261,45 +254,6 @@ jobs:
           path: /tmp/nemoclaw-traces/
           if-no-files-found: ignore
 
-  # ── Strict Chat Completions tool-call probe E2E ──────────────
-  # Coverage guard for #4537. Local Ollama onboarding requires structured
-  # Chat Completions tool calls; this hermetic mock asserts the bounded
-  # validation payload and retry behavior without GPU/Ollama infrastructure.
-  strict-tool-call-probe-e2e:
-    needs: select_regression_jobs
-    if: >-
-      github.repository == 'NVIDIA/NemoClaw' &&
-      needs.select_regression_jobs.outputs.strict_tool_call_probe == 'true'
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-    timeout-minutes: 15
-    steps:
-      - name: Checkout
-        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
-        with:
-          persist-credentials: false
-
-      - name: Setup Node
-        uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6
-        with:
-          node-version: "22"
-
-      - name: Run strict tool-call probe E2E test
-        env:
-          NEMOCLAW_TEST_NO_SLEEP: "1"
-        run: bash test/e2e/test-strict-tool-call-probe.sh
-
-      - name: Upload strict tool-call probe logs on failure
-        if: failure()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        with:
-          name: strict-tool-call-probe-logs
-          path: |
-            /tmp/nemoclaw-e2e-strict-tool-call-probe.log
-            /tmp/nemoclaw-e2e-strict-tool-call-probe-node.log
-          if-no-files-found: ignore
-
   # ── Gateway drift preflight E2E ─────────────────────────────
   # Coverage guard for #3399 / #3423. A stale OpenShell gateway image can
   # make sandbox-state RPCs fail with protobuf invalid-wire decode errors.

test/e2e-script-workflow.test.ts

@@ -76,7 +76,6 @@ const LEGACY_E2E_SHELL_ALLOWLIST = [
   "test/e2e/test-snapshot-commands.sh",
   "test/e2e/test-spark-install.sh",
   "test/e2e/test-state-backup-restore.sh",
-  "test/e2e/test-strict-tool-call-probe.sh",
   "test/e2e/test-telegram-injection.sh",
   "test/e2e/test-token-rotation.sh",
   "test/e2e/test-tunnel-lifecycle.sh",

test/e2e/test-strict-tool-call-probe.sh → test/fixtures/strict-tool-call-probe-driver.ts

@@ -1,60 +1,46 @@
-#!/usr/bin/env bash
-# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Coverage guard for #4537. The Local Ollama onboarding path is the only
-# current caller that requires strict Chat Completions tool calls. This
-# hermetic E2E exercises that validation path against an OpenAI-compatible
-# mock endpoint so payload-shape and retry regressions do not require a GPU
-# Ollama runner to catch.
-
-set -euo pipefail
-
-LOG_FILE="/tmp/nemoclaw-e2e-strict-tool-call-probe.log"
-exec > >(tee "$LOG_FILE") 2>&1
-
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m'
-
-pass() { echo -e "${GREEN}[PASS]${NC} $1"; }
-info() { echo -e "${YELLOW}[INFO]${NC} $1"; }
-diag() { echo -e "${YELLOW}[DIAG]${NC} $1"; }
-fail() {
-  echo -e "${RED}[FAIL]${NC} $1" >&2
-  diag "strict tool-call probe log tail:"
-  tail -120 "$LOG_FILE" 2>/dev/null || true
-  exit 1
-}
-
-SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
-REPO_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
-cd "$REPO_ROOT"
-
-info "Preparing CLI build"
-if [ ! -d node_modules ]; then
-  npm ci --ignore-scripts
-fi
-npm run build:cli
-
-info "Running strict Chat Completions tool-call probe against a hermetic mock"
-set +e
-NEMOCLAW_TEST_NO_SLEEP=1 node <<'NODE' 2>&1 | tee /tmp/nemoclaw-e2e-strict-tool-call-probe-node.log
-const assert = require("node:assert/strict");
-const { spawn, spawnSync } = require("node:child_process");
-const fs = require("node:fs");
-const os = require("node:os");
-const path = require("node:path");
+// @ts-nocheck
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+//
+// Process-level driver for the Local Ollama strict Chat Completions
+// tool-call probe. Loaded by test/strict-tool-call-probe.test.ts via
+// `tsx <driver>`; not picked up by Vitest's discovery (lives under
+// test/fixtures/, which is excluded from the test glob).
+//
+// Mirrors the inline `node -e` block from the retired
+// test/e2e/test-strict-tool-call-probe.sh, retained here so the
+// caller-level behavior under test stays identical to production
+// runtime conditions (subprocess curl probes, real env propagation,
+// no Vitest worker shims). Refs #4537, #4349, #5098, #5119.
+//
+// CWD must be the repo root; cli build artifacts under dist/ are required.
+//
+// Authored as TypeScript (rather than .cjs) per the codebase-growth
+// guardrail forbidding newly added .js/.cjs/.mjs files. Body is JS-shaped
+// because the embedded `node -e` strings must remain plain CommonJS for
+// the spawned children, and the dist/lib/* targets are CJS modules.
+// `@ts-nocheck` keeps the surface unchanged from the retired bash heredoc.
+
+import assert from "node:assert/strict";
+import { spawn, spawnSync } from "node:child_process";
+import fs from "node:fs";
+import { createRequire } from "node:module";
+import os from "node:os";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
 
 process.env.NEMOCLAW_TEST_NO_SLEEP = "1";
 process.env.NO_PROXY = [process.env.NO_PROXY, "127.0.0.1", "localhost"].filter(Boolean).join(",");
 process.env.no_proxy = [process.env.no_proxy, "127.0.0.1", "localhost"].filter(Boolean).join(",");
 
-const {
-  createInferenceSelectionValidationHelpers,
-} = require("./dist/lib/onboard/inference-selection-validation");
-const localInference = require("./dist/lib/inference/local");
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+const REPO_ROOT = path.resolve(__dirname, "..", "..");
+const requireFromHere = createRequire(import.meta.url);
+const { createInferenceSelectionValidationHelpers } = requireFromHere(
+  path.join(REPO_ROOT, "dist", "lib", "onboard", "inference-selection-validation"),
+);
+const localInference = requireFromHere(path.join(REPO_ROOT, "dist", "lib", "inference", "local"));
 
 function assertStrictPayload(payload) {
   assert.equal(payload.model, "mock-tool-model");
@@ -240,6 +226,9 @@ async function withMockEndpoint(mode, exercise) {
 
 function runOnboardingCallerAgainstMock(endpoint) {
   const port = new URL(endpoint).port;
+  // The child runs with cwd=REPO_ROOT (set via spawnSync below) so its
+  // `./dist/...` requires resolve consistently regardless of how this
+  // driver was launched.
   const childScript = String.raw`
 const assert = require("node:assert/strict");
 
@@ -308,7 +297,7 @@ console.error = (...args) => lines.push(args.join(" "));
 `;
 
   const result = spawnSync(process.execPath, ["-e", childScript], {
-    cwd: process.cwd(),
+    cwd: REPO_ROOT,
     encoding: "utf8",
     env: { ...process.env, NEMOCLAW_OLLAMA_PORT: port },
     timeout: 15000,
@@ -339,7 +328,9 @@ console.error = (...args) => lines.push(args.join(" "));
     assert.equal(requests[0].method, "POST");
     assert.equal(requests[0].url, "/v1/chat/completions");
     assertStrictPayload(requests[0].body);
-    console.log("[PASS] Local Ollama onboarding caller enforces strict Chat Completions validation");
+    console.log(
+      "[PASS] Local Ollama onboarding caller enforces strict Chat Completions validation",
+    );
   });
 
   await withMockEndpoint("transient-502", async (endpoint, readRequests) => {
@@ -366,12 +357,3 @@ console.error = (...args) => lines.push(args.join(" "));
   console.error(error && error.stack ? error.stack : error);
   process.exit(1);
 });
-NODE
-NODE_EXIT=$?
-set -e
-
-if [ "$NODE_EXIT" -ne 0 ]; then
-  fail "strict Chat Completions tool-call probe harness failed"
-fi
-
-pass "strict Chat Completions tool-call probe E2E passed"

test/regression-e2e-workflow.test.ts

@@ -36,4 +36,15 @@ describe("Regression E2E workflow contract", () => {
     expect(selectorScript).not.toContain("docker-unreachable-gateway-start-e2e");
     expect(selectorScript).not.toContain("docker_unreachable_gateway_start");
   });
+
+  it("does not advertise or select the retired strict-tool-call-probe lane", () => {
+    const jobsDescription = workflow.on?.workflow_dispatch?.inputs?.jobs?.description ?? "";
+    const selectorScript =
+      workflow.jobs?.select_regression_jobs?.steps?.find((step) => step.id === "select")?.run ?? "";
+
+    expect(jobsDescription).not.toContain("strict-tool-call-probe-e2e");
+    expect(Object.keys(workflow.jobs ?? {})).not.toContain("strict-tool-call-probe-e2e");
+    expect(selectorScript).not.toContain("strict-tool-call-probe-e2e");
+    expect(selectorScript).not.toContain("strict_tool_call_probe");
+  });
 });

test/strict-tool-call-probe.test.ts

@@ -0,0 +1,87 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+import assert from "node:assert/strict";
+import { spawnSync } from "node:child_process";
+import fs from "node:fs";
+import path from "node:path";
+import { describe, it } from "vitest";
+
+import { testTimeoutOptions } from "./helpers/timeouts";
+
+// Coverage guard for #4537. The Local Ollama onboarding path is the only
+// current caller that requires strict Chat Completions tool calls. This
+// hermetic, caller-level Vitest test exercises that validation path against
+// an OpenAI-compatible mock endpoint so payload-shape and retry regressions
+// do not require a GPU/Ollama runner to catch.
+//
+// Replaces test/e2e/test-strict-tool-call-probe.sh per #5119 retirement
+// pattern: caller-level mock-driven probes belong in test/, not in live E2E
+// scenario/fixture surfaces or the regression-e2e bash workflow. Refs #5098, #4349.
+//
+// Why subprocess: the validation path drives `curl` via spawnSync with a
+// tight process timeout. Driving the entire scenario set through a fresh
+// `tsx <driver>` child mirrors the legacy script (and #5119's
+// onboard-gateway-docker-unreachable.test.ts) and keeps the behavior under
+// test identical to production runtime conditions — bypassing Vitest's
+// worker pool, fetch shim, and signal handling, all of which can interfere
+// with the in-process curl subprocess used by validateOpenAiLikeSelection.
+//
+// The driver is `.ts` (executed via tsx) rather than `.cjs` per the
+// codebase-growth guardrail that forbids newly added .js/.cjs/.mjs files.
+
+const REPO_ROOT = path.join(import.meta.dirname, "..");
+const TSX = path.join(REPO_ROOT, "node_modules", ".bin", "tsx");
+const DRIVER = path.join(import.meta.dirname, "fixtures", "strict-tool-call-probe-driver.ts");
+const REQUIRED_DIST_MODULES = [
+  path.join(REPO_ROOT, "dist", "lib", "onboard", "inference-selection-validation.js"),
+  path.join(REPO_ROOT, "dist", "lib", "inference", "local.js"),
+];
+
+const EXPECTED_PASS_MARKERS = [
+  "[PASS] strict validation succeeds with structured tool_calls",
+  "[PASS] Local Ollama onboarding caller enforces strict Chat Completions validation",
+  "[PASS] strict validation retries a transient 502 and keeps bounded payloads",
+  "[PASS] strict validation fails closed when no structured tool_call is returned",
+];
+
+describe("strict Chat Completions tool-call probe (#4537)", () => {
+  it(
+    "validates Local Ollama strict tool-call enforcement against a hermetic mock",
+    testTimeoutOptions(120_000),
+    () => {
+      const missingDistModules = REQUIRED_DIST_MODULES.filter(
+        (modulePath) => !fs.existsSync(modulePath),
+      );
+      assert.deepEqual(
+        missingDistModules,
+        [],
+        `strict tool-call probe requires built CLI artifacts; run npm run build:cli first. Missing:\n${missingDistModules.join("\n")}`,
+      );
+
+      const result = spawnSync(TSX, [DRIVER], {
+        cwd: REPO_ROOT,
+        encoding: "utf8",
+        env: { ...process.env, NEMOCLAW_TEST_NO_SLEEP: "1" },
+        timeout: 110_000,
+        // Inherit stderr for diagnostic visibility on failure; capture stdout
+        // to assert the [PASS] markers below.
+        stdio: ["ignore", "pipe", "inherit"],
+      });
+
+      const stdout = result.stdout ?? "";
+      assert.equal(
+        result.status,
+        0,
+        `strict tool-call probe driver exited with ${result.status}; stdout:\n${stdout}`,
+      );
+
+      for (const marker of EXPECTED_PASS_MARKERS) {
+        assert.ok(
+          stdout.includes(marker),
+          `missing pass marker ${JSON.stringify(marker)} in driver stdout:\n${stdout}`,
+        );
+      }
+    },
+  );
+});