Spaces:
Running
Running
File size: 10,638 Bytes
4d7e460 c8e1a6c 4d7e460 c8e1a6c 4d7e460 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 | """Eval: Does the LLM invoke headroom_retrieve when summaries are present?
The REAL test — it's not enough for the LLM to know something is missing.
It must actually call the tool to fetch it.
Compares:
- WITH summary: LLM sees "2 failed, 1 error" → should call headroom_retrieve
- WITHOUT summary: LLM sees "[90 items compressed]" → likely does NOT call tool
Requires: ANTHROPIC_API_KEY in environment or .env file.
Run: python -m pytest tests/test_compression_summary_tool_eval.py -v -s
"""
from __future__ import annotations
import json
import os
import pytest
from tests._dotenv import autouse_apply_env, load_env_overrides
_env_overrides = load_env_overrides()
ANTHROPIC_KEY = os.environ.get("ANTHROPIC_API_KEY") or _env_overrides.get("ANTHROPIC_API_KEY", "")
apply_dotenv = autouse_apply_env(_env_overrides)
pytestmark = pytest.mark.skipif(
not ANTHROPIC_KEY,
reason="ANTHROPIC_API_KEY not set — skipping integration tests",
)
# The headroom_retrieve tool definition (same as what CCR injects)
HEADROOM_RETRIEVE_TOOL = {
"name": "headroom_retrieve",
"description": (
"Retrieve original uncompressed content from Headroom's compression cache. "
"Use this when you need more details from compressed data. "
"You can pass a query to search within the compressed content."
),
"input_schema": {
"type": "object",
"properties": {
"hash": {
"type": "string",
"description": "The hash key from the compression marker",
},
"query": {
"type": "string",
"description": "Optional search query to find specific items within the compressed data",
},
},
"required": ["hash"],
},
}
def _call_claude_with_tools(messages: list[dict], tools: list[dict], max_tokens: int = 300) -> dict:
"""Make a real Anthropic API call with tool use."""
import httpx
resp = httpx.post(
"https://api.anthropic.com/v1/messages",
headers={
"X-Api-Key": ANTHROPIC_KEY,
"anthropic-version": "2023-06-01",
"Content-Type": "application/json",
},
json={
"model": "claude-sonnet-4-5-20250929",
"max_tokens": max_tokens,
"messages": messages,
"tools": tools,
},
timeout=30,
)
return resp.json()
def _make_test_results(n: int = 100) -> list[dict]:
"""Test suite output with hidden failures in the compressed portion."""
results = []
for i in range(n):
result = {
"test_name": f"test_module_{i // 10}.test_case_{i}",
"status": "passed",
"duration_ms": 50 + i * 3,
}
if i == 42:
result["status"] = "failed"
result["error"] = "AssertionError: expected 200, got 401 in auth_middleware"
result["test_name"] = "test_auth.test_login_expired_token"
if i == 67:
result["status"] = "failed"
result["error"] = "TimeoutError: database pool exhausted after 30s"
result["test_name"] = "test_database.test_concurrent_connections"
if i == 88:
result["status"] = "error"
result["error"] = "ImportError: cannot import 'NewFeature'"
result["test_name"] = "test_features.test_new_feature_integration"
results.append(result)
return results
def _has_tool_use(response: dict) -> bool:
"""Check if the response contains a tool_use block."""
for block in response.get("content", []):
if block.get("type") == "tool_use":
return True
return False
def _get_tool_calls(response: dict) -> list[dict]:
"""Extract all tool_use blocks from response."""
calls = []
for block in response.get("content", []):
if block.get("type") == "tool_use":
calls.append(
{
"name": block.get("name"),
"input": block.get("input", {}),
}
)
return calls
class TestToolInvocationWithSummary:
"""The real eval: does the LLM call headroom_retrieve?"""
def test_with_summary_triggers_tool_call(self):
"""WITH compression summary → LLM should call headroom_retrieve."""
test_results = _make_test_results(100)
kept = test_results[:10] # All passing
from headroom.transforms.compression_summary import summarize_dropped_items
summary = summarize_dropped_items(test_results, kept)
compressed = json.dumps(kept, indent=2)
compressed += (
f"\n[90 items compressed to 10. Omitted: {summary}."
f' Retrieve specific items: headroom_retrieve(hash="ccr_test_abc123", query="your search")]'
)
messages = [
{
"role": "user",
"content": (
"Here are the test results from our CI pipeline:\n\n"
f"{compressed}\n\n"
"Tell me about any test failures. What went wrong?"
),
},
]
resp = _call_claude_with_tools(messages, [HEADROOM_RETRIEVE_TOOL])
tool_calls = _get_tool_calls(resp)
stop_reason = resp.get("stop_reason", "")
print(f"\n Summary: {summary}")
print(f" Stop reason: {stop_reason}")
print(f" Tool calls: {tool_calls}")
# With a summary showing failures, the LLM SHOULD call the tool
if stop_reason == "tool_use":
assert len(tool_calls) > 0
call = tool_calls[0]
assert call["name"] == "headroom_retrieve"
assert call["input"].get("hash") == "ccr_test_abc123"
# The query should be about failures/errors
query = call["input"].get("query", "").lower()
print(f" Query used: {query}")
has_relevant_query = any(
term in query for term in ["fail", "error", "issue", "problem", "broken", "test"]
)
assert has_relevant_query, f"Tool was called but query isn't relevant: {query}"
print(" RESULT: LLM invoked headroom_retrieve with relevant query ✓")
else:
# LLM responded with text — check if it at least mentions the failures
text = ""
for block in resp.get("content", []):
if block.get("type") == "text":
text += block.get("text", "")
print(f" LLM text response: {text[:200]}")
# It's acceptable if the LLM mentions it WANTS to retrieve
mentions_retrieval = any(
term in text.lower()
for term in ["retrieve", "headroom_retrieve", "fetch", "see more", "compressed"]
)
print(f" Mentions retrieval: {mentions_retrieval}")
def test_without_summary_baseline(self):
"""WITHOUT compression summary → LLM likely does NOT call tool."""
test_results = _make_test_results(100)
kept = test_results[:10] # All passing
compressed = json.dumps(kept, indent=2)
compressed += "\n[90 items compressed to 10. Retrieve more: hash=ccr_test_abc123]"
messages = [
{
"role": "user",
"content": (
"Here are the test results from our CI pipeline:\n\n"
f"{compressed}\n\n"
"Tell me about any test failures. What went wrong?"
),
},
]
resp = _call_claude_with_tools(messages, [HEADROOM_RETRIEVE_TOOL])
tool_calls = _get_tool_calls(resp)
stop_reason = resp.get("stop_reason", "")
print(f"\n Stop reason: {stop_reason}")
print(f" Tool calls: {tool_calls}")
if stop_reason == "tool_use":
call = tool_calls[0]
print(f" Query used: {call['input'].get('query', 'none')}")
print(" RESULT: LLM DID invoke tool (may check proactively)")
else:
text = ""
for block in resp.get("content", []):
if block.get("type") == "text":
text += block.get("text", "")
print(f" LLM text response: {text[:200]}")
print(" RESULT: LLM did NOT invoke tool — assumed all tests passed")
def test_code_summary_triggers_retrieval(self):
"""Code compression summary → LLM should retrieve specific function."""
compressed_code = '''class PaymentProcessor:
"""Processes payments via Stripe."""
def __init__(self, api_key: str):
# [2 lines omitted]
pass
def charge(self, amount: float, currency: str, token: str) -> dict:
# [8 lines omitted]
pass
def refund(self, charge_id: str, amount: float = None) -> dict:
# [3 lines omitted]
pass
def get_balance(self) -> float:
# [2 lines omitted]
pass
# [180 tokens compressed. removed: def charge (12 lines), def refund (6 lines). Retrieve full code: headroom_retrieve(hash="ccr_code_xyz", query="function name")]'''
messages = [
{
"role": "user",
"content": (
"Here's the payment processor code:\n\n"
f"```python\n{compressed_code}\n```\n\n"
"There's a bug in the retry logic for failed charges. "
"Can you find and fix it?"
),
},
]
resp = _call_claude_with_tools(messages, [HEADROOM_RETRIEVE_TOOL])
tool_calls = _get_tool_calls(resp)
stop_reason = resp.get("stop_reason", "")
print(f"\n Stop reason: {stop_reason}")
print(f" Tool calls: {tool_calls}")
if stop_reason == "tool_use":
call = tool_calls[0]
assert call["name"] == "headroom_retrieve"
query = call["input"].get("query", "").lower()
print(f" Query: {query}")
# Should be asking for the charge function specifically
has_charge = any(term in query for term in ["charge", "retry", "payment", "stripe"])
print(f" Targets charge/retry: {has_charge}")
print(" RESULT: LLM invoked tool to get the charge() implementation ✓")
else:
text = ""
for block in resp.get("content", []):
if block.get("type") == "text":
text += block.get("text", "")
print(f" LLM text: {text[:200]}")
print(" RESULT: LLM did NOT invoke tool")
|