fix: report health-check failures instead of skipping them

.github/workflows/health-checks.yml

@@ -37,7 +37,7 @@ jobs:
     name: ${{ matrix.provider }} end to end
     runs-on: ubuntu-latest
     # A provider outage should not hide the state of the other two.
-    continue-on-error: true
+    timeout-minutes: 50
     strategy:
       fail-fast: false
       matrix:
@@ -93,23 +93,20 @@ jobs:
           make health-check \
             PROVIDER_CONFIG="$PROVIDER_CONFIG" \
             HEALTH_LOCALE="$HEALTH_LOCALE" \
-            | tee /tmp/health.log
+            HEALTH_SUMMARY_FILE=/tmp/usersim-health-summary.md
 
       - name: Summary
         if: always()
         env:
           PROVIDER: ${{ matrix.provider }}
           LOCALE: ${{ matrix.locale }}
-          SECRET_NAME: ${{ matrix.secret_name }}
         run: |
           {
             echo "### $PROVIDER ($LOCALE)"
             echo
-            if grep -q '^SKIP' /tmp/health.log; then
-              echo "Skipped: repository secret \`$SECRET_NAME\` is not set."
+            if test -f /tmp/usersim-health-summary.md; then
+              cat /tmp/usersim-health-summary.md
             else
-              echo '```'
-              tail -20 /tmp/health.log || echo "(no output captured)"
-              echo '```'
+              echo "**failure** (setup): health check did not produce an outcome."
             fi
           } >> "$GITHUB_STEP_SUMMARY"

CONTRIBUTING.md

@@ -233,7 +233,10 @@ make health-check PROVIDER_CONFIG=src/usersim/cli/models_openai.toml
 ```
 
 A provider whose API key you have not set is reported as skipped rather than
-failed, so you can run this with one key and check one provider.
+failed, so you can run this with one key and check one provider. Invalid
+configuration and execution errors fail the check. It loads `.env.local` before
+`.env`, preserving exported variables. Each CLI command has a 900-second timeout;
+set `HEALTH_TIMEOUT` to change it.
 
 It runs a real one-row `simulate`, `eval` and `report`. Please run it if you
 touched models, providers, prompts, scorers or the trajectory store. The

Makefile

@@ -228,45 +228,15 @@ PROVIDER_CONFIG ?= src/usersim/cli/models_default.toml
 HEALTH_LOCALE   ?= ja_JP
 HEALTH_PROBE    ?= sov_ai_dynamic
 
-# A provider you have no key for is skipped rather than failed. Without this
-# the target is red for everyone who has one key, and the scheduled workflow
-# is red until every secret is provisioned, which trains people to ignore it.
-define REQUIRE_KEYS_PY
-import sys
-from usersim.cli._models import load_models_config, required_api_key_env_vars
-config = sys.argv[1]
-missing = sorted(v for v in required_api_key_env_vars(load_models_config(config)) if not __import__("os").environ.get(v))
-if missing:
-    print(f"SKIP  {config}: {', '.join(missing)} not set")
-    sys.exit(1)
-endef
-export REQUIRE_KEYS_PY
+# Only missing credentials skip; configuration and execution errors fail.
+HEALTH_TIMEOUT ?= 900
+HEALTH_SUMMARY_FILE ?=
 
 health-check:  ## Live one-row simulate + eval + report against a provider
-	@set -eu; \
-	if ! $(UV) run python -c "$$REQUIRE_KEYS_PY" "$(PROVIDER_CONFIG)"; then exit 0; fi; \
-	OUT=$$(mktemp -d); \
-	trap 'rm -rf "$$OUT"' EXIT; \
-	echo "==> simulate  ($(HEALTH_LOCALE), $(PROVIDER_CONFIG))"; \
-	$(UV) run usersim simulate \
-		--locale $(HEALTH_LOCALE) --num-rows 1 --max-turns 2 \
-		--probe-mix "$(HEALTH_PROBE)=1.0" \
-		--models $(PROVIDER_CONFIG) \
-		--out "$$OUT/trajectories"; \
-	RUN=$$(basename $$(ls -d "$$OUT"/trajectories/run=* | head -1) | cut -d= -f2); \
-	echo "==> eval      (run $$RUN)"; \
-	$(UV) run usersim eval --run "$$RUN" \
-		--trajectories "$$OUT/trajectories" \
-		--models $(PROVIDER_CONFIG) \
-		--out "$$OUT/evaluations"; \
-	echo "==> report    (run $$RUN)"; \
-	$(UV) run usersim report --run "$$RUN" \
-		--trajectories "$$OUT/trajectories" \
-		--evaluations "$$OUT/evaluations" \
-		--out "$$OUT/report"; \
-	test -f "$$OUT/report/run=$$RUN/index.html" \
-		|| { echo "report produced no index.html"; exit 1; }; \
-	echo "✓ simulate, eval and report all completed"
+	@$(UV) run python scripts/check_health.py \
+		--models "$(PROVIDER_CONFIG)" --locale "$(HEALTH_LOCALE)" \
+		--probe "$(HEALTH_PROBE)" --timeout "$(HEALTH_TIMEOUT)" \
+		$(if $(HEALTH_SUMMARY_FILE),--summary-file "$(HEALTH_SUMMARY_FILE)")
 
 check-extensions:  ## Install an out-of-tree package and prove all six seams work
 	@set -eu; \

scripts/check_health.py

@@ -0,0 +1,131 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Run the live CLI pipeline, skipping only absent provider credentials."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+import sys
+import tempfile
+from pathlib import Path
+
+
+def preflight(models: Path) -> list[str]:
+    # Keep imports within the reported boundary: a broken installation fails,
+    # rather than masquerading as an unconfigured provider.
+    import dotenv  # noqa: F401 -- required by load_local_env, which tolerates its absence
+
+    from usersim.cli._env import load_local_env, missing_keys
+    from usersim.cli._models import (
+        REQUIRED_EVALUATOR_ALIASES,
+        REQUIRED_SIMULATOR_ALIASES,
+        assistant_replays_reasoning,
+        load_models_config,
+        require_aliases,
+        required_api_key_env_vars,
+        to_model_configs,
+    )
+
+    load_local_env()
+    config = load_models_config(models)
+    require_aliases(
+        config,
+        required=REQUIRED_SIMULATOR_ALIASES + REQUIRED_EVALUATOR_ALIASES,
+        context="health check",
+    )
+    assistant_replays_reasoning(config, store_reasoning=True)
+    # Validate model parameters through the same offline conversion as the CLI.
+    to_model_configs(config)
+    return sorted(missing_keys(required_api_key_env_vars(config)))
+
+
+def run_stage(stage: str, arguments: list[str], timeout: float) -> None:
+    print(f"==> {stage}", flush=True)
+    subprocess.run([sys.executable, "-m", "usersim.cli", stage, *arguments], check=True, timeout=timeout)
+
+
+def pipeline(args: argparse.Namespace, output: Path, outcome: dict[str, str]) -> None:
+    trajectories = str(output / "trajectories")
+    evaluations = str(output / "evaluations")
+    outcome["stage"] = "simulate"
+    run_stage(
+        "simulate",
+        [
+            "--locale",
+            args.locale,
+            "--num-rows",
+            "1",
+            "--max-turns",
+            "2",
+            "--probe-mix",
+            f"{args.probe}=1.0",
+            "--models",
+            str(args.models),
+            "--out",
+            trajectories,
+        ],
+        args.timeout,
+    )
+    runs = sorted(path for path in Path(trajectories).glob("run=*") if path.is_dir())
+    if len(runs) != 1:
+        raise RuntimeError("simulate must produce exactly one run directory")
+    run = runs[0].name.removeprefix("run=")
+    outcome["stage"] = "eval"
+    run_stage(
+        "eval",
+        ["--run", run, "--trajectories", trajectories, "--models", str(args.models), "--out", evaluations],
+        args.timeout,
+    )
+    outcome["stage"] = "report"
+    run_stage(
+        "report",
+        ["--run", run, "--trajectories", trajectories, "--evaluations", evaluations, "--out", str(output / "report")],
+        args.timeout,
+    )
+    if not (output / "report" / f"run={run}" / "index.html").is_file():
+        raise RuntimeError("report produced no index.html")
+
+
+def error_detail(error: Exception) -> str:
+    # Avoid printing subprocess arguments or credential-bearing exception text.
+    if isinstance(error, subprocess.CalledProcessError):
+        return f"command exited with status {error.returncode}"
+    if isinstance(error, subprocess.TimeoutExpired):
+        return f"command exceeded {error.timeout:g} seconds"
+    if isinstance(error, RuntimeError):
+        return str(error)
+    return f"{type(error).__name__}; check configuration, dependencies and installation"
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--models", type=Path, required=True)
+    parser.add_argument("--locale", default="ja_JP")
+    parser.add_argument("--probe", default="sov_ai_dynamic")
+    parser.add_argument("--timeout", type=float, default=900, help="maximum seconds per CLI command (default: 900)")
+    parser.add_argument("--summary-file", type=Path)
+    args = parser.parse_args(argv)
+    outcome = {"status": "failure", "stage": "preflight", "detail": ""}
+    try:
+        if args.timeout <= 0 or not args.timeout < float("inf"):
+            raise ValueError("timeout must be finite and positive")
+        missing = preflight(args.models)
+        if missing:
+            outcome.update(status="skip", detail=f"missing credentials: {', '.join(missing)}")
+        else:
+            with tempfile.TemporaryDirectory(prefix="usersim-health-") as temporary:
+                pipeline(args, Path(temporary), outcome)
+            outcome.update(status="pass", stage="complete", detail="simulate, eval and report completed")
+    except Exception as error:
+        outcome["detail"] = error_detail(error)
+    print(json.dumps(outcome), flush=True)
+    if args.summary_file:
+        args.summary_file.write_text(f"**{outcome['status']}** ({outcome['stage']}): {outcome['detail']}\n")
+    return int(outcome["status"] == "failure")
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())