feat(01-03): integration tests for Slack flow, rate limiting, and agent persona

- tests/unit/test_ratelimit.py: 11 tests for Redis token bucket (CHAN-05)
  - allows requests under limit, rejects 31st request
  - per-tenant isolation, per-channel isolation
  - TTL key expiry and window reset
- tests/integration/test_slack_flow.py: 15 tests for end-to-end Slack flow (CHAN-02)
  - normalization: bot token stripped, channel=slack, thread_id set
  - @mention: placeholder posted in-thread, Celery dispatched with placeholder_ts
  - DM flow: same pipeline triggered for channel_type=im
  - bot messages silently ignored (no infinite loop)
  - unknown workspace_id silently ignored
  - duplicate events (Slack retries) skipped via idempotency
- tests/integration/test_agent_persona.py: 15 tests for persona in prompts (AGNT-01)
  - system prompt contains name, role, persona, AI transparency clause
  - model_preference forwarded to LLM pool
  - full messages array: [system, user] structure verified
- tests/integration/test_ratelimit.py: 4 tests for rate limit integration
  - over-limit -> ephemeral rejection posted
  - over-limit -> Celery NOT dispatched, placeholder NOT posted
  - within-limit -> no rejection
  - ephemeral message includes actionable retry hint
All 45 tests pass
This commit is contained in:
2026-03-23 10:32:48 -06:00
parent 6f30705e1a
commit 74326dfc3d
4 changed files with 1127 additions and 0 deletions

View File

@@ -0,0 +1,307 @@
"""
Integration tests for agent persona reflection in LLM system prompts (AGNT-01).
Tests verify:
1. The system prompt contains the agent's name, role, and persona
2. The AI transparency clause is always present
3. model_preference from the agent config is passed to the LLM pool
4. The full message array (system + user) is correctly structured
These tests mock the LLM pool HTTP call — no real LLM API keys required.
They test the orchestrator -> agent builder -> runner chain in isolation.
"""
from __future__ import annotations
import uuid
from unittest.mock import AsyncMock, MagicMock, patch
import pytest
from orchestrator.agents.builder import build_messages, build_system_prompt
class _MockAgent:
"""Minimal mock of the Agent ORM model for unit testing the builder."""
def __init__(
self,
name: str,
role: str,
persona: str,
system_prompt: str = "",
model_preference: str = "quality",
) -> None:
self.id = uuid.uuid4()
self.tenant_id = uuid.uuid4()
self.name = name
self.role = role
self.persona = persona
self.system_prompt = system_prompt
self.model_preference = model_preference
self.is_active = True
class TestAgentPersonaInSystemPrompt:
"""AGNT-01: Agent identity and persona must appear in the system prompt."""
def test_agent_name_in_system_prompt(self) -> None:
"""System prompt must contain 'Your name is {agent.name}'."""
agent = _MockAgent(name="Mara", role="Customer Support", persona="Professional and empathetic")
prompt = build_system_prompt(agent)
assert "Mara" in prompt
assert "Your name is Mara" in prompt
def test_agent_role_in_system_prompt(self) -> None:
"""System prompt must contain the agent's role."""
agent = _MockAgent(name="Mara", role="Customer Support", persona="Professional and empathetic")
prompt = build_system_prompt(agent)
assert "Customer Support" in prompt
assert "Your role is Customer Support" in prompt
def test_agent_persona_in_system_prompt(self) -> None:
"""System prompt must include the agent's persona text."""
agent = _MockAgent(
name="Mara",
role="Customer Support",
persona="Professional and empathetic",
)
prompt = build_system_prompt(agent)
assert "Professional and empathetic" in prompt
def test_ai_transparency_clause_always_present(self) -> None:
"""
The AI transparency clause must be present in every system prompt,
regardless of agent configuration.
Agents must acknowledge they are AIs when directly asked.
"""
agent = _MockAgent(name="Mara", role="Support", persona="")
prompt = build_system_prompt(agent)
# The clause uses the word "AI" — verify it's unconditionally injected
assert "AI" in prompt or "artificial intelligence" in prompt.lower()
# Verify the specific phrase from builder.py
assert "you are an AI" in prompt.lower() or "you are an ai" in prompt.lower()
def test_ai_transparency_present_even_with_empty_persona(self) -> None:
"""Transparency clause must appear even when persona is empty."""
agent = _MockAgent(name="Bot", role="Assistant", persona="")
prompt = build_system_prompt(agent)
assert "AI" in prompt
def test_custom_system_prompt_included(self) -> None:
"""If agent has a base system_prompt, it must appear in the output."""
agent = _MockAgent(
name="Mara",
role="Support",
persona="Helpful",
system_prompt="Always be concise.",
)
prompt = build_system_prompt(agent)
assert "Always be concise." in prompt
def test_full_persona_customer_support_scenario(self) -> None:
"""
Full system prompt for a 'Mara' customer support agent must contain
all required elements.
"""
agent = _MockAgent(
name="Mara",
role="Customer Support",
persona="Professional, empathetic, solution-oriented.",
)
prompt = build_system_prompt(agent)
assert "Mara" in prompt
assert "Customer Support" in prompt
assert "Professional, empathetic, solution-oriented." in prompt
assert "AI" in prompt # Transparency clause
def test_name_and_role_on_same_line(self) -> None:
"""Name and role must appear together in the identity sentence."""
agent = _MockAgent(name="Atlas", role="DevOps Engineer", persona="")
prompt = build_system_prompt(agent)
assert "Your name is Atlas. Your role is DevOps Engineer." in prompt
class TestAgentPersonaInMessages:
"""Verify the full messages array structure passed to the LLM pool."""
def test_messages_has_system_message_first(self) -> None:
"""The first message must be the system message."""
agent = _MockAgent(name="Mara", role="Support", persona="Helpful")
prompt = build_system_prompt(agent)
messages = build_messages(system_prompt=prompt, user_message="Hello")
assert messages[0]["role"] == "system"
assert messages[0]["content"] == prompt
def test_messages_has_user_message_last(self) -> None:
"""The last message must be the user message."""
agent = _MockAgent(name="Mara", role="Support", persona="Helpful")
prompt = build_system_prompt(agent)
user_text = "Can you help with my order?"
messages = build_messages(system_prompt=prompt, user_message=user_text)
assert messages[-1]["role"] == "user"
assert messages[-1]["content"] == user_text
def test_messages_has_exactly_two_entries_no_history(self) -> None:
"""Without history, messages must have exactly [system, user]."""
agent = _MockAgent(name="Mara", role="Support", persona="Helpful")
prompt = build_system_prompt(agent)
messages = build_messages(system_prompt=prompt, user_message="Hi")
assert len(messages) == 2
def test_messages_includes_history_in_order(self) -> None:
"""Conversation history must appear between system and user messages."""
agent = _MockAgent(name="Mara", role="Support", persona="Helpful")
prompt = build_system_prompt(agent)
history = [
{"role": "user", "content": "Previous question"},
{"role": "assistant", "content": "Previous answer"},
]
messages = build_messages(system_prompt=prompt, user_message="Follow-up", history=history)
# Structure: system, history[0], history[1], user
assert len(messages) == 4
assert messages[1] == history[0]
assert messages[2] == history[1]
assert messages[-1]["role"] == "user"
class TestModelPreferencePassthrough:
"""Verify model_preference is passed correctly to the LLM pool."""
async def test_model_preference_passed_to_llm_pool(self) -> None:
"""
The agent's model_preference must be forwarded as the 'model' field
in the LLM pool /complete request payload.
"""
from orchestrator.agents.runner import run_agent
from shared.models.message import ChannelType, KonstructMessage, MessageContent, SenderInfo
from datetime import datetime, timezone
agent = _MockAgent(
name="Mara",
role="Customer Support",
persona="Professional and empathetic",
model_preference="quality",
)
msg = KonstructMessage(
tenant_id=str(agent.tenant_id),
channel=ChannelType.SLACK,
channel_metadata={"workspace_id": "T-TEST"},
sender=SenderInfo(user_id="U1", display_name="Test User"),
content=MessageContent(text="Hello Mara"),
timestamp=datetime.now(tz=timezone.utc),
)
captured_payloads: list[dict] = []
async def mock_post_response(*args, **kwargs):
payload = kwargs.get("json", {})
captured_payloads.append(payload)
mock_resp = MagicMock()
mock_resp.status_code = 200
mock_resp.json.return_value = {"content": "Hello from Mara!", "model": "quality"}
return mock_resp
with patch("httpx.AsyncClient") as mock_http_class:
mock_http_instance = AsyncMock()
mock_http_instance.__aenter__ = AsyncMock(return_value=mock_http_instance)
mock_http_instance.__aexit__ = AsyncMock(return_value=False)
mock_http_instance.post = AsyncMock(side_effect=mock_post_response)
mock_http_class.return_value = mock_http_instance
result = await run_agent(msg, agent)
assert len(captured_payloads) == 1
payload = captured_payloads[0]
assert payload["model"] == "quality"
async def test_llm_response_returned_as_string(self) -> None:
"""run_agent must return the LLM response as a plain string."""
from orchestrator.agents.runner import run_agent
from shared.models.message import ChannelType, KonstructMessage, MessageContent, SenderInfo
from datetime import datetime, timezone
agent = _MockAgent(
name="Mara",
role="Support",
persona="Helpful",
model_preference="fast",
)
msg = KonstructMessage(
tenant_id=str(agent.tenant_id),
channel=ChannelType.SLACK,
channel_metadata={},
sender=SenderInfo(user_id="U1", display_name="Test"),
content=MessageContent(text="What is 2+2?"),
timestamp=datetime.now(tz=timezone.utc),
)
with patch("httpx.AsyncClient") as mock_http_class:
mock_http_instance = AsyncMock()
mock_http_instance.__aenter__ = AsyncMock(return_value=mock_http_instance)
mock_http_instance.__aexit__ = AsyncMock(return_value=False)
mock_response = MagicMock()
mock_response.status_code = 200
mock_response.json.return_value = {"content": "The answer is 4.", "model": "fast"}
mock_http_instance.post = AsyncMock(return_value=mock_response)
mock_http_class.return_value = mock_http_instance
result = await run_agent(msg, agent)
assert isinstance(result, str)
assert result == "The answer is 4."
async def test_system_prompt_forwarded_to_llm_pool(self) -> None:
"""
The system prompt (including persona + AI clause) must be the first
message in the array sent to the LLM pool.
"""
from orchestrator.agents.runner import run_agent
from shared.models.message import ChannelType, KonstructMessage, MessageContent, SenderInfo
from datetime import datetime, timezone
agent = _MockAgent(
name="Mara",
role="Customer Support",
persona="Professional and empathetic",
)
msg = KonstructMessage(
tenant_id=str(agent.tenant_id),
channel=ChannelType.SLACK,
channel_metadata={},
sender=SenderInfo(user_id="U1", display_name="Test"),
content=MessageContent(text="hi"),
timestamp=datetime.now(tz=timezone.utc),
)
captured_messages: list = []
async def capture_request(*args, **kwargs):
captured_messages.extend(kwargs.get("json", {}).get("messages", []))
mock_resp = MagicMock()
mock_resp.status_code = 200
mock_resp.json.return_value = {"content": "Hello!", "model": "quality"}
return mock_resp
with patch("httpx.AsyncClient") as mock_http_class:
mock_http_instance = AsyncMock()
mock_http_instance.__aenter__ = AsyncMock(return_value=mock_http_instance)
mock_http_instance.__aexit__ = AsyncMock(return_value=False)
mock_http_instance.post = AsyncMock(side_effect=capture_request)
mock_http_class.return_value = mock_http_instance
await run_agent(msg, agent)
assert len(captured_messages) >= 2
system_msg = captured_messages[0]
assert system_msg["role"] == "system"
# System prompt must contain all persona elements
assert "Mara" in system_msg["content"]
assert "Customer Support" in system_msg["content"]
assert "Professional and empathetic" in system_msg["content"]
assert "AI" in system_msg["content"] # Transparency clause