Files
crewAI/lib/crewai/tests/hooks/test_llm_hooks.py
Lucas Gomide 7d21283630 feat: add generic interception-hook dispatcher (#6516)
* feat: add generic interception-hook dispatcher

Introduces `crewai/hooks/dispatch.py` as a single engine behind every
interception point: a hook receives a typed context, may mutate or replace
its `payload`, or raise `HookAborted(reason, source)` to stop the operation.
The full `InterceptionPoint` catalog is frozen from day zero, with global and
contextvar-scoped registries, an `@on` decorator, a no-op fast path, and a
`HookDispatchedEvent` for telemetry. The four existing `before/after_llm_call`
and `before/after_tool_call` hooks become adapters over the dispatcher, so the
legacy dialect and `return False` semantics keep working unchanged while
gaining the new contract.

* fix: harden interception dispatcher against review findings

Corrects several dispatcher edge cases surfaced in review. `_default_reducer`
now reports a modification only when a `payload` is actually applied, the
`agents=` filter falls back to `agent_role` for contexts without an `agent`
object, and `unregister` resolves the filter wrapper stashed by `on` so a
filtered hook can be removed. The tool-hook runners honor the executing
agent's `verbose` flag instead of silently swallowing hook errors, and the
ReAct tool path now runs `POST_TOOL_CALL` on blocked calls to match the
native paths. Also adds abort-telemetry coverage and replaces the flaky
absolute no-op timing budget with a relative one.

* fix: honor scoped hooks on direct llm calls and register @on crew methods

Direct agent-less LLM calls short-circuited on the empty global hook list,
so hooks registered only for the current `scoped_hooks()` context never
ran; the direct-call helpers now defer to `dispatch`, which resolves
scoped hooks behind its own no-op fast path. `CrewBase` likewise only
scanned the legacy `is_*_hook` markers, so `@on(InterceptionPoint.X)`
methods were silently dropped — it now registers them on the dispatcher
with filters applied and `self` bound. Also tightens result typing across
the tool-call seams so `mypy` stays green.

* refactor: scope InterceptionPoint to the points this layer wires

The dispatcher only fires the model- and tool-call boundaries, so
`InterceptionPoint` now lists just those four rather than the full future
catalog. New points are introduced alongside the seams that dispatch them,
keeping every layer free of enum members with no live consumer. The
dispatcher unit tests that borrowed unused points as generic examples are
remapped onto the four kept points.

* test: pin per-hook fail-open at the LLM and tool seams

The dispatcher swallows a hook's exception per hook rather than around the
whole loop, so one buggy hook no longer silently skips every hook registered
after it. These seam-level tests pin that behavior through
`_setup_before_llm_call_hooks` and `run_before/after_tool_call_hooks`, and
confirm an intentional `return False` block still short-circuits later hooks.

* fix: run execution-scoped hooks on the agent executor model seams

`_setup_before/after_llm_call_hooks` only ran the executor's snapshot
hook lists, so hooks registered via `scoped_hooks()` never fired on
`PRE/POST_MODEL_CALL` during normal agent execution, while the tool
seams (which go through `dispatch`) merged them. The seams now append
the current scope's hooks after the snapshot via `get_scoped_hooks`,
matching dispatch's global-then-scoped ordering, and a scoped-only
registration no longer short-circuits the seam.
2026-07-14 10:27:34 -04:00

673 lines
24 KiB
Python

"""Unit tests for LLM hooks functionality."""
from __future__ import annotations
from unittest.mock import Mock
from crewai.hooks import (
clear_all_llm_call_hooks,
unregister_after_llm_call_hook,
unregister_before_llm_call_hook,
)
from crewai.hooks.llm_hooks import (
LLMCallHookContext,
get_after_llm_call_hooks,
get_before_llm_call_hooks,
register_after_llm_call_hook,
register_before_llm_call_hook,
)
import pytest
@pytest.fixture
def mock_executor():
"""Create a mock executor for testing."""
executor = Mock()
executor.messages = [{"role": "system", "content": "Test message"}]
executor.agent = Mock(role="Test Agent")
executor.task = Mock(description="Test Task")
executor.crew = Mock()
executor.llm = Mock()
executor.iterations = 0
return executor
@pytest.fixture(autouse=True)
def clear_hooks():
"""Clear global hooks before and after each test."""
from crewai.hooks import llm_hooks
original_before = llm_hooks._before_llm_call_hooks.copy()
original_after = llm_hooks._after_llm_call_hooks.copy()
llm_hooks._before_llm_call_hooks.clear()
llm_hooks._after_llm_call_hooks.clear()
yield
llm_hooks._before_llm_call_hooks.clear()
llm_hooks._after_llm_call_hooks.clear()
llm_hooks._before_llm_call_hooks.extend(original_before)
llm_hooks._after_llm_call_hooks.extend(original_after)
class TestLLMCallHookContext:
"""Test LLMCallHookContext initialization and attributes."""
def test_context_initialization(self, mock_executor):
"""Test that context is initialized correctly with executor."""
context = LLMCallHookContext(executor=mock_executor)
assert context.executor == mock_executor
assert context.messages == mock_executor.messages
assert context.agent == mock_executor.agent
assert context.task == mock_executor.task
assert context.crew == mock_executor.crew
assert context.llm == mock_executor.llm
assert context.iterations == mock_executor.iterations
assert context.response is None
def test_context_with_response(self, mock_executor):
"""Test that context includes response when provided."""
test_response = "Test LLM response"
context = LLMCallHookContext(executor=mock_executor, response=test_response)
assert context.response == test_response
def test_messages_are_mutable_reference(self, mock_executor):
"""Test that modifying context.messages modifies executor.messages."""
context = LLMCallHookContext(executor=mock_executor)
new_message = {"role": "user", "content": "New message"}
context.messages.append(new_message)
assert new_message in mock_executor.messages
assert len(mock_executor.messages) == 2
class TestBeforeLLMCallHooks:
"""Test before_llm_call hook registration and execution."""
def test_register_before_hook(self):
"""Test that before hooks are registered correctly."""
def test_hook(context):
pass
register_before_llm_call_hook(test_hook)
hooks = get_before_llm_call_hooks()
assert len(hooks) == 1
assert hooks[0] == test_hook
def test_multiple_before_hooks(self):
"""Test that multiple before hooks can be registered."""
def hook1(context):
pass
def hook2(context):
pass
register_before_llm_call_hook(hook1)
register_before_llm_call_hook(hook2)
hooks = get_before_llm_call_hooks()
assert len(hooks) == 2
assert hook1 in hooks
assert hook2 in hooks
def test_before_hook_can_modify_messages(self, mock_executor):
"""Test that before hooks can modify messages in-place."""
def add_message_hook(context):
context.messages.append({"role": "system", "content": "Added by hook"})
context = LLMCallHookContext(executor=mock_executor)
add_message_hook(context)
assert len(context.messages) == 2
assert context.messages[1]["content"] == "Added by hook"
def test_get_before_hooks_returns_copy(self):
"""Test that get_before_llm_call_hooks returns a copy."""
def test_hook(context):
pass
register_before_llm_call_hook(test_hook)
hooks1 = get_before_llm_call_hooks()
hooks2 = get_before_llm_call_hooks()
assert hooks1 == hooks2
assert hooks1 is not hooks2
class TestAfterLLMCallHooks:
"""Test after_llm_call hook registration and execution."""
def test_register_after_hook(self):
"""Test that after hooks are registered correctly."""
def test_hook(context):
return None
register_after_llm_call_hook(test_hook)
hooks = get_after_llm_call_hooks()
assert len(hooks) == 1
assert hooks[0] == test_hook
def test_multiple_after_hooks(self):
"""Test that multiple after hooks can be registered."""
def hook1(context):
return None
def hook2(context):
return None
register_after_llm_call_hook(hook1)
register_after_llm_call_hook(hook2)
hooks = get_after_llm_call_hooks()
assert len(hooks) == 2
assert hook1 in hooks
assert hook2 in hooks
def test_after_hook_can_modify_response(self, mock_executor):
"""Test that after hooks can modify the response."""
original_response = "Original response"
def modify_response_hook(context):
if context.response:
return context.response.replace("Original", "Modified")
return None
context = LLMCallHookContext(executor=mock_executor, response=original_response)
modified = modify_response_hook(context)
assert modified == "Modified response"
def test_after_hook_returns_none_keeps_original(self, mock_executor):
"""Test that returning None keeps the original response."""
original_response = "Original response"
def no_change_hook(context):
return None
context = LLMCallHookContext(executor=mock_executor, response=original_response)
result = no_change_hook(context)
assert result is None
assert context.response == original_response
def test_get_after_hooks_returns_copy(self):
"""Test that get_after_llm_call_hooks returns a copy."""
def test_hook(context):
return None
register_after_llm_call_hook(test_hook)
hooks1 = get_after_llm_call_hooks()
hooks2 = get_after_llm_call_hooks()
assert hooks1 == hooks2
assert hooks1 is not hooks2
class TestLLMHooksIntegration:
"""Test integration scenarios with multiple hooks."""
def test_multiple_before_hooks_execute_in_order(self, mock_executor):
"""Test that multiple before hooks execute in registration order."""
execution_order = []
def hook1(context):
execution_order.append(1)
def hook2(context):
execution_order.append(2)
def hook3(context):
execution_order.append(3)
register_before_llm_call_hook(hook1)
register_before_llm_call_hook(hook2)
register_before_llm_call_hook(hook3)
context = LLMCallHookContext(executor=mock_executor)
hooks = get_before_llm_call_hooks()
for hook in hooks:
hook(context)
assert execution_order == [1, 2, 3]
def test_multiple_after_hooks_chain_modifications(self, mock_executor):
"""Test that multiple after hooks can chain modifications."""
def hook1(context):
if context.response:
return context.response + " [hook1]"
return None
def hook2(context):
if context.response:
return context.response + " [hook2]"
return None
register_after_llm_call_hook(hook1)
register_after_llm_call_hook(hook2)
context = LLMCallHookContext(executor=mock_executor, response="Original")
hooks = get_after_llm_call_hooks()
result = context.response
for hook in hooks:
context.response = result
modified = hook(context)
if modified is not None:
result = modified
assert result == "Original [hook1] [hook2]"
def test_after_hooks_do_not_clobber_native_tool_call_responses(
self, mock_executor
):
"""A registered after hook must not break native tool execution.
Regression for crewAIInc/crewAI#6529: `_setup_after_llm_call_hooks`
stringified structured tool-call payloads, so the executor treated the
raw tool call as the final answer and never executed the tool. Non-str,
non-BaseModel responses now pass through untouched; hooks still fire on
textual responses.
"""
from crewai.utilities.agent_utils import _setup_after_llm_call_hooks
observed = []
def observer(context):
observed.append(context.response)
return None
register_after_llm_call_hook(observer)
mock_executor.after_llm_call_hooks = get_after_llm_call_hooks()
tool_calls = [Mock()] # structured native tool-call payload
result = _setup_after_llm_call_hooks(
mock_executor, tool_calls, printer=Mock(), verbose=False
)
assert result is tool_calls
text = _setup_after_llm_call_hooks(
mock_executor, "final answer", printer=Mock(), verbose=False
)
assert text == "final answer"
assert observed == ["final answer"]
def test_unregister_before_hook(self):
"""Test that before hooks can be unregistered."""
def test_hook(context):
pass
register_before_llm_call_hook(test_hook)
unregister_before_llm_call_hook(test_hook)
hooks = get_before_llm_call_hooks()
assert len(hooks) == 0
def test_unregister_after_hook(self):
"""Test that after hooks can be unregistered."""
def test_hook(context):
return None
register_after_llm_call_hook(test_hook)
unregister_after_llm_call_hook(test_hook)
hooks = get_after_llm_call_hooks()
assert len(hooks) == 0
def test_clear_all_llm_call_hooks(self):
"""Test that all llm call hooks can be cleared."""
def test_hook(context):
pass
register_before_llm_call_hook(test_hook)
register_after_llm_call_hook(test_hook)
clear_all_llm_call_hooks()
hooks = get_before_llm_call_hooks()
assert len(hooks) == 0
def test_raising_before_hook_does_not_skip_later_hooks(self, mock_executor):
"""Fail-open is per-hook: a crashing hook must not disable its neighbors.
Regression guard for the dispatcher migration: previously the
``except Exception`` wrapped the whole hook loop, so a raising hook
silently skipped every hook registered after it. Now swallowing is
per-hook — later hooks still run and the LLM call still proceeds.
"""
from crewai.utilities.agent_utils import _setup_before_llm_call_hooks
ran: list[str] = []
def crashing_hook(context):
ran.append("crashing")
raise ValueError("bug in user hook")
def later_hook(context):
ran.append("later")
register_before_llm_call_hook(crashing_hook)
register_before_llm_call_hook(later_hook)
mock_executor.before_llm_call_hooks = get_before_llm_call_hooks()
proceed = _setup_before_llm_call_hooks(
mock_executor, printer=Mock(), verbose=False
)
assert ran == ["crashing", "later"]
assert proceed is True
def test_scoped_hooks_fire_on_agent_executor_llm_seams(self, mock_executor):
"""register_scoped hooks must run on the executor model seams.
Regression: `_setup_before/after_llm_call_hooks` only ran the
executor's snapshot lists, so execution-scoped hooks never fired on
PRE/POST_MODEL_CALL during normal agent execution (while tool seams,
which go through `dispatch`, merged them). Scoped hooks run after the
snapshot, matching dispatch's global-then-scoped ordering.
"""
from crewai.hooks import InterceptionPoint
from crewai.hooks.dispatch import register_scoped, scoped_hooks
from crewai.utilities.agent_utils import (
_setup_after_llm_call_hooks,
_setup_before_llm_call_hooks,
)
order: list[str] = []
def snapshot_hook(context):
order.append("snapshot")
mock_executor.before_llm_call_hooks = [snapshot_hook]
mock_executor.after_llm_call_hooks = []
with scoped_hooks():
register_scoped(
InterceptionPoint.PRE_MODEL_CALL,
lambda ctx: order.append("scoped_pre"),
)
register_scoped(
InterceptionPoint.POST_MODEL_CALL,
lambda ctx: order.append("scoped_post"),
)
proceed = _setup_before_llm_call_hooks(
mock_executor, printer=Mock(), verbose=False
)
answer = _setup_after_llm_call_hooks(
mock_executor, "answer", printer=Mock(), verbose=False
)
assert order == ["snapshot", "scoped_pre", "scoped_post"]
assert proceed is True
assert answer == "answer"
def test_intentional_block_still_short_circuits_later_hooks(self, mock_executor):
"""A hook returning False blocks the call and skips later hooks (unchanged)."""
from crewai.utilities.agent_utils import _setup_before_llm_call_hooks
ran: list[str] = []
def blocking_hook(context):
ran.append("blocking")
return False
def later_hook(context):
ran.append("later")
register_before_llm_call_hook(blocking_hook)
register_before_llm_call_hook(later_hook)
mock_executor.before_llm_call_hooks = get_before_llm_call_hooks()
proceed = _setup_before_llm_call_hooks(
mock_executor, printer=Mock(), verbose=False
)
assert ran == ["blocking"]
assert proceed is False
@pytest.mark.vcr()
def test_lite_agent_hooks_integration_with_real_llm(self):
"""Test that LiteAgent executes before/after LLM call hooks and prints messages correctly."""
import os
from crewai.lite_agent import LiteAgent
if not os.environ.get("OPENAI_API_KEY"):
pytest.skip("OPENAI_API_KEY not set - skipping real LLM test")
hook_calls = {"before": [], "after": []}
def before_llm_call_hook(context: LLMCallHookContext) -> bool:
"""Log and verify before hook execution."""
print(f"\n[BEFORE HOOK] Agent: {context.agent.role if context.agent else 'None'}")
print(f"[BEFORE HOOK] Iterations: {context.iterations}")
print(f"[BEFORE HOOK] Message count: {len(context.messages)}")
print(f"[BEFORE HOOK] Messages: {context.messages}")
hook_calls["before"].append({
"iterations": context.iterations,
"message_count": len(context.messages),
"has_task": context.task is not None,
"has_crew": context.crew is not None,
})
return True
def after_llm_call_hook(context: LLMCallHookContext) -> str | None:
"""Log and verify after hook execution."""
print(f"\n[AFTER HOOK] Agent: {context.agent.role if context.agent else 'None'}")
print(f"[AFTER HOOK] Iterations: {context.iterations}")
print(f"[AFTER HOOK] Response: {context.response[:100] if context.response else 'None'}...")
print(f"[AFTER HOOK] Final message count: {len(context.messages)}")
hook_calls["after"].append({
"iterations": context.iterations,
"has_response": context.response is not None,
"response_length": len(context.response) if context.response else 0,
})
if context.response:
return f"[HOOKED] {context.response}"
return None
register_before_llm_call_hook(before_llm_call_hook)
register_after_llm_call_hook(after_llm_call_hook)
try:
lite_agent = LiteAgent(
role="Test Assistant",
goal="Answer questions briefly",
backstory="You are a helpful test assistant",
verbose=True,
)
assert len(lite_agent.before_llm_call_hooks) > 0, "Before hooks not loaded"
assert len(lite_agent.after_llm_call_hooks) > 0, "After hooks not loaded"
result = lite_agent.kickoff("Say 'Hello World' and nothing else")
assert len(hook_calls["before"]) > 0, "Before hook was never called"
assert len(hook_calls["after"]) > 0, "After hook was never called"
# LiteAgent doesn't have task/crew context, unlike agents in CrewBase
before_call = hook_calls["before"][0]
assert before_call["has_task"] is False, "Task should be None for LiteAgent in flows"
assert before_call["has_crew"] is False, "Crew should be None for LiteAgent in flows"
assert before_call["message_count"] > 0, "Should have messages"
after_call = hook_calls["after"][0]
assert after_call["has_response"] is True, "After hook should have response"
assert after_call["response_length"] > 0, "Response should not be empty"
# Note: The hook modifies the raw LLM response, but LiteAgent then parses it
# to extract the "Final Answer" portion. We check the messages to see the modification.
assert len(result.messages) > 2, "Should have assistant message in messages"
last_message = result.messages[-1]
assert last_message["role"] == "assistant", "Last message should be from assistant"
assert "[HOOKED]" in last_message["content"], "Hook should have modified the assistant message"
finally:
unregister_before_llm_call_hook(before_llm_call_hook)
unregister_after_llm_call_hook(after_llm_call_hook)
@pytest.mark.vcr()
def test_direct_llm_call_hooks_integration(self):
"""Test that hooks work for direct llm.call() without agents."""
import os
from crewai.llm import LLM
if not os.environ.get("OPENAI_API_KEY"):
pytest.skip("OPENAI_API_KEY not set - skipping real LLM test")
hook_calls = {"before": [], "after": []}
def before_hook(context: LLMCallHookContext) -> bool:
"""Log and verify before hook execution."""
print(f"\n[BEFORE HOOK] Agent: {context.agent}")
print(f"[BEFORE HOOK] Task: {context.task}")
print(f"[BEFORE HOOK] Crew: {context.crew}")
print(f"[BEFORE HOOK] LLM: {context.llm}")
print(f"[BEFORE HOOK] Iterations: {context.iterations}")
print(f"[BEFORE HOOK] Message count: {len(context.messages)}")
hook_calls["before"].append({
"agent": context.agent,
"task": context.task,
"crew": context.crew,
"llm": context.llm is not None,
"message_count": len(context.messages),
})
return True
def after_hook(context: LLMCallHookContext) -> str | None:
"""Log and verify after hook execution."""
print(f"\n[AFTER HOOK] Agent: {context.agent}")
print(f"[AFTER HOOK] Response: {context.response[:100] if context.response else 'None'}...")
hook_calls["after"].append({
"has_response": context.response is not None,
"response_length": len(context.response) if context.response else 0,
})
if context.response:
return f"[HOOKED] {context.response}"
return None
register_before_llm_call_hook(before_hook)
register_after_llm_call_hook(after_hook)
try:
llm = LLM(model="gpt-4o-mini")
result = llm.call([{"role": "user", "content": "Say hello"}])
print(f"\n[TEST] Final result: {result}")
assert len(hook_calls["before"]) > 0, "Before hook was never called"
assert len(hook_calls["after"]) > 0, "After hook was never called"
before_call = hook_calls["before"][0]
assert before_call["agent"] is None, "Agent should be None for direct LLM calls"
assert before_call["task"] is None, "Task should be None for direct LLM calls"
assert before_call["crew"] is None, "Crew should be None for direct LLM calls"
assert before_call["llm"] is True, "LLM should be present"
assert before_call["message_count"] > 0, "Should have messages"
after_call = hook_calls["after"][0]
assert after_call["has_response"] is True, "After hook should have response"
assert after_call["response_length"] > 0, "Response should not be empty"
assert "[HOOKED]" in result, "Response should be modified by after hook"
finally:
unregister_before_llm_call_hook(before_hook)
unregister_after_llm_call_hook(after_hook)
class TestDirectLLMScopedHooks:
"""Direct (agent-less) LLM calls must honor execution-scoped hooks.
Regression: the direct-call helpers used to short-circuit when the global
hook list was empty, so hooks registered only for the current
``scoped_hooks()`` context never ran on this path.
"""
@staticmethod
def _stub_llm():
from crewai.llms.base_llm import BaseLLM
class _StubLLM(BaseLLM):
def call(self, *args: object, **kwargs: object) -> str:
return ""
return _StubLLM(model="stub")
def test_scoped_before_hook_runs_on_direct_call(self):
from crewai.hooks import InterceptionPoint
from crewai.hooks.dispatch import register_scoped, scoped_hooks
llm = self._stub_llm()
seen: list[int] = []
with scoped_hooks():
register_scoped(
InterceptionPoint.PRE_MODEL_CALL,
lambda ctx: seen.append(len(ctx.messages)),
)
proceed = llm._invoke_before_llm_call_hooks(
[{"role": "user", "content": "hi"}], from_agent=None
)
assert proceed is True
assert seen == [1]
def test_scoped_before_hook_can_block_direct_call(self):
from crewai.hooks import InterceptionPoint
from crewai.hooks.dispatch import HookAborted, register_scoped, scoped_hooks
llm = self._stub_llm()
def block(ctx: LLMCallHookContext) -> None:
raise HookAborted(reason="blocked by scoped hook")
with scoped_hooks():
register_scoped(InterceptionPoint.PRE_MODEL_CALL, block)
proceed = llm._invoke_before_llm_call_hooks(
[{"role": "user", "content": "hi"}], from_agent=None
)
assert proceed is False
def test_scoped_after_hook_modifies_direct_response(self):
from crewai.hooks import InterceptionPoint
from crewai.hooks.dispatch import register_scoped, scoped_hooks
llm = self._stub_llm()
def redact(ctx: LLMCallHookContext) -> str:
return ctx.response.replace("SECRET", "[REDACTED]")
with scoped_hooks():
register_scoped(InterceptionPoint.POST_MODEL_CALL, redact)
result = llm._invoke_after_llm_call_hooks(
[{"role": "user", "content": "hi"}],
"contains SECRET",
from_agent=None,
)
assert result == "contains [REDACTED]"