fix: fix test

Signed-off-by: Nick Goncharenko <[email protected]>

Author: ngoncharenko

docs/evaluator/agent-eval/evaluate-deployed-agent.mdx

@@ -131,7 +131,7 @@ from nemo_evaluator_sdk.agent_inference import make_agent_inference_fn
 agent = GenericAgent(
     name="hermes-agent",
     url="http://127.0.0.1:8642/v1/responses",
-    api_key_secret=SecretRef("API_SERVER_KEY"),
+    api_key_secret=SecretRef("HERMES_TOKEN"),
     body={
         "model": "hermes-agent",
         "input": "{{ instruction }}",
@@ -153,7 +153,7 @@ evaluator = AgentEvaluator(
 )
 ```
 
-The SDK sends the bearer token from `API_SERVER_KEY` env variable, preserves the raw stream as evidence, and
+The SDK sends the bearer token from the `HERMES_TOKEN` environment variable, preserves the raw stream as evidence, and
 passes the extracted output text to the task's metrics. The Hermes translator reads the terminal
 `response.completed` payload, validates completion, and records user/agent steps plus token usage in
 ATIF-v1.7.

packages/nemo_evaluator_sdk/examples/hermes/example.py

@@ -154,7 +154,7 @@ def __call__(
         return AgentStreamTranslation(trajectory=trajectory.model_dump(mode="python", exclude_none=True))
 
 
-def _require_api_server_key() -> None:
+def _require_hermes_token() -> None:
     if os.getenv(HERMES_TOKEN_ENV_NAME):
         return
     raise RuntimeError(
@@ -170,7 +170,7 @@ async def evaluate(
     client: httpx.AsyncClient | None = None,
 ) -> AgentEvalResult:
     """Run the Hermes streaming evaluation."""
-    _require_api_server_key()
+    _require_hermes_token()
     evaluator = AgentEvaluator(
         agent_inference_fn_factory=partial(
             make_agent_inference_fn,
@@ -184,7 +184,7 @@ async def evaluate(
     task = AgentEvalTask(
         id="hermes-streaming",
         intent="Return the requested phrase over SSE.",
-        inputs={"instruction": f"Reply with exactly: {EXPECTED_TEXT}"},
+        inputs={"instruction": f"Reply with exactly: {EXPECTED_TEXT}."},
         reference={"expected": EXPECTED_TEXT},
         metrics=[KeywordMatchMetric()],
     )

packages/nemo_evaluator_sdk/src/nemo_evaluator_sdk/agent_eval/docs/architecture.md

@@ -468,7 +468,7 @@ Hermes supplies domain knowledge through a translator while using a
 target = GenericAgent(
     url="http://127.0.0.1:8642/v1/responses",
     name="hermes-agent",
-    api_key_secret=SecretRef("API_SERVER_KEY"),
+    api_key_secret=SecretRef("HERMES_TOKEN"),
     body={
         "model": "hermes-agent",
         "input": "{{ instruction }}",

packages/nemo_evaluator_sdk/tests/agent_eval/test_hermes_example.py

@@ -56,7 +56,7 @@ def handle(request: httpx.Request) -> httpx.Response:
             headers={"content-type": "text/event-stream"},
         )
 
-    monkeypatch.setenv("API_SERVER_KEY", "test-token")
+    monkeypatch.setenv(hermes.HERMES_TOKEN_ENV_NAME, "test-token")
     async with httpx.AsyncClient(transport=httpx.MockTransport(handle)) as client:
         result = await hermes.evaluate(agent_url="http://hermes.test/v1/responses", client=client)
 
@@ -204,7 +204,7 @@ async def test_incomplete_stream_records_translation_error(monkeypatch: pytest.M
 event: response.output_text.done
 data: {"type":"response.output_text.done","text":"streaming works"}
 """
-    monkeypatch.setenv("API_SERVER_KEY", "test-token")
+    monkeypatch.setenv(hermes.HERMES_TOKEN_ENV_NAME, "test-token")
     transport = httpx.MockTransport(
         lambda request: httpx.Response(200, content=incomplete, headers={"content-type": "text/event-stream"})
     )
@@ -221,20 +221,20 @@ async def test_incomplete_stream_records_translation_error(monkeypatch: pytest.M
 
 
 @pytest.mark.asyncio
-async def test_missing_api_server_key_fails_before_request(monkeypatch: pytest.MonkeyPatch) -> None:
+async def test_missing_hermes_token_fails_before_request(monkeypatch: pytest.MonkeyPatch) -> None:
     request_count = 0
 
     def handle(request: httpx.Request) -> httpx.Response:
         nonlocal request_count
         request_count += 1
         return httpx.Response(500, request=request)
 
-    monkeypatch.delenv(hermes.API_SERVER_KEY_ENV_NAME, raising=False)
+    monkeypatch.delenv(hermes.HERMES_TOKEN_ENV_NAME, raising=False)
     async with httpx.AsyncClient(transport=httpx.MockTransport(handle)) as client:
         with pytest.raises(RuntimeError) as exc_info:
             await hermes.evaluate(agent_url="http://hermes.test/v1/responses", client=client)
 
     message = str(exc_info.value)
     assert request_count == 0
-    assert "export API_SERVER_KEY=<your-api-server-key>" in message
+    assert "export HERMES_TOKEN=<your-hermes-token>" in message
     assert hermes.HERMES_API_SERVER_DOCS_URL in message