test: add pytest suite + CI workflow (21 tests)

tests/test_chat_mocked.py

@@ -0,0 +1,100 @@
+"""Tests for /api/chat/{agent_id} with a mocked LLM.
+
+B2 regression: the user's message must reach the LLM (was broken when history
+slice dropped the new turn-message). Test by capturing the messages list
+passed to ``llm.chat`` and asserting the last item is the user's input.
+"""
+import pytest
+
+from llm import LLMResponse
+
+
+pytestmark = pytest.mark.asyncio
+
+
+def _fake_llm_response(content: str = "FAKE_REPLY", tool_calls=None, usage=None) -> LLMResponse:
+    return LLMResponse(content=content, tool_calls=tool_calls or [], usage=usage or {
+        "prompt_tokens": 1,
+        "completion_tokens": 1,
+        "total_tokens": 2,
+        "finish_reason": "stop",
+    })
+
+
+async def test_chat_empty_message_returns_422(app_client, auth_headers, seed_agent):
+    """POST /api/chat with empty message -> 422 (B6 min_length=1)."""
+    resp = await app_client.post(
+        "/api/chat/test_agent",
+        json={"message": ""},
+        headers=auth_headers,
+    )
+    assert resp.status_code == 422, resp.text
+
+
+async def test_chat_message_over_limit_422(app_client, auth_headers, seed_agent):
+    """POST /api/chat with 40000-char message -> 422 (B6 max_length=32000)."""
+    resp = await app_client.post(
+        "/api/chat/test_agent",
+        json={"message": "x" * 40000},
+        headers=auth_headers,
+    )
+    assert resp.status_code == 422, resp.text
+
+
+async def test_chat_with_mocked_llm_returns_assistant(app_client, auth_headers, seed_agent, monkeypatch, stub_mcp, clean_db):
+    """Stub llm.chat -> 200 with assistant message; messages table has 2 rows (user + assistant)."""
+    async def fake_chat(messages, tools=None, model=None, temperature=0.7, max_tokens=2000, **_kw):
+        return _fake_llm_response(content="FAKE_REPLY")
+
+    monkeypatch.setattr("agent.chat", fake_chat)
+
+    resp = await app_client.post(
+        "/api/chat/test_agent",
+        json={"message": "hello world"},
+        headers=auth_headers,
+    )
+    assert resp.status_code == 200, resp.text
+    body = resp.json()
+    assert body["content"] == "FAKE_REPLY"
+    assert "conversation_id" in body
+    conv_id = body["conversation_id"]
+
+    # Verify messages table: 1 user + 1 assistant = 2 rows
+    async with clean_db.execute(
+        "SELECT role, content FROM messages WHERE conversation_id = ? ORDER BY id ASC",
+        (conv_id,),
+    ) as cur:
+        rows = await cur.fetchall()
+    assert len(rows) == 2, f"expected 2 rows (user+assistant), got {len(rows)}"
+    assert rows[0][0] == "user"
+    assert rows[0][1] == "hello world"
+    assert rows[1][0] == "assistant"
+    assert rows[1][1] == "FAKE_REPLY"
+
+
+async def test_chat_user_message_in_prompt(app_client, auth_headers, seed_agent, monkeypatch, stub_mcp):
+    """B2 regression: capture messages list passed to llm.chat, assert last item == user input."""
+    captured = {}
+
+    async def capturing_chat(messages, tools=None, model=None, temperature=0.7, max_tokens=2000, **_kw):
+        captured["messages"] = list(messages)
+        return _fake_llm_response(content="FAKE_REPLY")
+
+    monkeypatch.setattr("agent.chat", capturing_chat)
+
+    user_input = "what is the meaning of life, the universe, and everything?"
+    resp = await app_client.post(
+        "/api/chat/test_agent",
+        json={"message": user_input},
+        headers=auth_headers,
+    )
+    assert resp.status_code == 200, resp.text
+
+    assert "messages" in captured, "llm.chat was not invoked"
+    msgs = captured["messages"]
+    assert len(msgs) >= 2
+    # First message must be the system prompt
+    assert msgs[0]["role"] == "system"
+    # Last message must be the user's input (B2 regression)
+    assert msgs[-1]["role"] == "user"
+    assert msgs[-1]["content"] == user_input