File size: 5,233 Bytes
b5b9c2e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 | """Tests for context token tracking in run_agent.py's usage extraction.
The context counter (status bar) must show the TOTAL prompt tokens including
Anthropic's cached portions. This is an integration test for the token
extraction in run_conversation(), not the ContextCompressor itself (which
is tested in tests/agent/test_context_compressor.py).
"""
import sys
import types
from types import SimpleNamespace
sys.modules.setdefault("fire", types.SimpleNamespace(Fire=lambda *a, **k: None))
sys.modules.setdefault("firecrawl", types.SimpleNamespace(Firecrawl=object))
sys.modules.setdefault("fal_client", types.SimpleNamespace())
import run_agent
def _patch_bootstrap(monkeypatch):
monkeypatch.setattr(run_agent, "get_tool_definitions", lambda **kwargs: [{
"type": "function",
"function": {"name": "t", "description": "t", "parameters": {"type": "object", "properties": {}}},
}])
monkeypatch.setattr(run_agent, "check_toolset_requirements", lambda: {})
class _FakeAnthropicClient:
def close(self):
pass
class _FakeOpenAIClient:
"""Fake OpenAI client returned by mocked resolve_provider_client."""
api_key = "fake-codex-key"
base_url = "https://api.openai.com/v1"
_default_headers = None
def _make_agent(monkeypatch, api_mode, provider, response_fn):
_patch_bootstrap(monkeypatch)
if api_mode == "anthropic_messages":
monkeypatch.setattr("agent.anthropic_adapter.build_anthropic_client", lambda k, b=None: _FakeAnthropicClient())
if provider == "openai-codex":
monkeypatch.setattr(
"agent.auxiliary_client.resolve_provider_client",
lambda *a, **kw: (_FakeOpenAIClient(), "test-model"),
)
class _A(run_agent.AIAgent):
def __init__(self, *a, **kw):
kw.update(skip_context_files=True, skip_memory=True, max_iterations=4)
super().__init__(*a, **kw)
self._cleanup_task_resources = self._persist_session = lambda *a, **k: None
self._save_trajectory = self._save_session_log = lambda *a, **k: None
def run_conversation(self, msg, conversation_history=None, task_id=None):
self._interruptible_api_call = lambda kw: response_fn()
self._disable_streaming = True
return super().run_conversation(msg, conversation_history=conversation_history, task_id=task_id)
return _A(model="test-model", api_key="test-key", provider=provider, api_mode=api_mode)
def _anthropic_resp(input_tok, output_tok, cache_read=0, cache_creation=0):
usage_fields = {"input_tokens": input_tok, "output_tokens": output_tok}
if cache_read:
usage_fields["cache_read_input_tokens"] = cache_read
if cache_creation:
usage_fields["cache_creation_input_tokens"] = cache_creation
return SimpleNamespace(
content=[SimpleNamespace(type="text", text="ok")],
stop_reason="end_turn",
usage=SimpleNamespace(**usage_fields),
model="claude-sonnet-4-6",
)
# -- Anthropic: cached tokens must be included --
def test_anthropic_cache_read_and_creation_added(monkeypatch):
agent = _make_agent(monkeypatch, "anthropic_messages", "anthropic",
lambda: _anthropic_resp(3, 10, cache_read=15000, cache_creation=2000))
agent.run_conversation("hi")
assert agent.context_compressor.last_prompt_tokens == 17003 # 3+15000+2000
assert agent.session_prompt_tokens == 17003
def test_anthropic_no_cache_fields(monkeypatch):
agent = _make_agent(monkeypatch, "anthropic_messages", "anthropic",
lambda: _anthropic_resp(500, 20))
agent.run_conversation("hi")
assert agent.context_compressor.last_prompt_tokens == 500
def test_anthropic_cache_read_only(monkeypatch):
agent = _make_agent(monkeypatch, "anthropic_messages", "anthropic",
lambda: _anthropic_resp(5, 15, cache_read=17666, cache_creation=15))
agent.run_conversation("hi")
assert agent.context_compressor.last_prompt_tokens == 17686 # 5+17666+15
# -- OpenAI: prompt_tokens already total --
def test_openai_prompt_tokens_unchanged(monkeypatch):
resp = lambda: SimpleNamespace(
choices=[SimpleNamespace(index=0, message=SimpleNamespace(
role="assistant", content="ok", tool_calls=None, reasoning_content=None,
), finish_reason="stop")],
usage=SimpleNamespace(prompt_tokens=5000, completion_tokens=100, total_tokens=5100),
model="gpt-4o",
)
agent = _make_agent(monkeypatch, "chat_completions", "openrouter", resp)
agent.run_conversation("hi")
assert agent.context_compressor.last_prompt_tokens == 5000
# -- Codex: no cache fields, getattr returns 0 --
def test_codex_no_cache_fields(monkeypatch):
resp = lambda: SimpleNamespace(
output=[SimpleNamespace(type="message", content=[SimpleNamespace(type="output_text", text="ok")])],
usage=SimpleNamespace(input_tokens=3000, output_tokens=50, total_tokens=3050),
status="completed", model="gpt-5-codex",
)
agent = _make_agent(monkeypatch, "codex_responses", "openai-codex", resp)
agent.run_conversation("hi")
assert agent.context_compressor.last_prompt_tokens == 3000
|