graff — a fast agentic coding harness in Zig: multi-provider, MCP, workflows, DGM evolution loop, TS/Python SDKs
Python
#!/usr/bin/env python3
"""Offline ACP thought-level option and next-request wire regression."""
import importlib.util
import sys
import tempfile
from pathlib import Path
REPO = Path(__file__).resolve().parents[1]
loader = importlib.util.spec_from_file_location("acp_load_test", REPO / "scripts/test-acp-session-load.py")
load_test = importlib.util.module_from_spec(loader)
loader.loader.exec_module(load_test)
Acp = load_test.Acp
ScriptedModel = load_test.ScriptedModel
def option(response):
options = response["result"]["configOptions"]
assert len(options) == 1, options
selected = options[0]
assert selected["id"] == selected["category"] == "thought_level", selected
assert selected["type"] == "select", selected
assert selected["name"] == "Thought Level", selected
return selected
def run(binary):
with tempfile.TemporaryDirectory(prefix="graff-acp-effort-") as temporary:
cwd = Path(temporary)
home = cwd / "home"
home.mkdir()
model = ScriptedModel([{"text": "first"}, {"text": "second"}])
port = model.start(0)
acp = Acp(binary, cwd, home, port)
try:
acp.request("initialize", {"protocolVersion": 1})
created = acp.request("session/new", {"cwd": str(cwd), "mcpServers": []})
sid = created["result"]["sessionId"]
initial = option(created)
assert initial["currentValue"] == "medium", initial
assert [item["value"] for item in initial["options"]] == ["low", "medium", "high", "xhigh", "max", "ultra"]
for params in (
{"sessionId": "wrong", "configId": "thought_level", "value": "high"},
{"sessionId": sid, "configId": "unknown", "value": "high"},
{"sessionId": sid, "configId": "thought_level", "value": "invalid"},
{"sessionId": sid, "configId": "thought_level", "value": True},
):
rejected = acp.request("session/set_config_option", params)
assert rejected["error"]["code"] == -32602, rejected
assert not model.requests, "a configuration request started inference"
selected = acp.request("session/set_config_option", {"sessionId": sid, "configId": "thought_level", "value": "high"})
assert option(selected)["currentValue"] == "high", selected
assert not model.requests, "selecting effort started inference"
prompt = lambda text: {"sessionId": sid, "prompt": [{"type": "text", "text": text}]}
first = acp.request("session/prompt", prompt("Say first."), 35)
assert first["result"]["stopReason"] == "end_turn", first
assert model.requests[0]["reasoning"]["effort"] == "high", model.requests[0]
before = len(model.requests)
start = len(acp.events)
changed = acp.request("session/prompt", prompt("/effort low"))
assert changed["result"]["stopReason"] == "end_turn", changed
assert len(model.requests) == before, "slash command started inference"
updates = [event["params"]["update"] for event in acp.events[start:] if event.get("method") == "session/update"]
changed_options = [update for update in updates if update.get("sessionUpdate") == "config_option_update"]
assert len(changed_options) == 1, updates
assert option({"result": changed_options[0]})["currentValue"] == "low", changed_options
second = acp.request("session/prompt", prompt("Say second."), 35)
assert second["result"]["stopReason"] == "end_turn", second
assert model.requests[1]["reasoning"]["effort"] == "low", model.requests[1]
finally:
acp.close()
model.stop()
print("ACP thought-level config, notifications and next-request effort: ok")
if __name__ == "__main__":
run(Path(sys.argv[1] if len(sys.argv) > 1 else REPO / "zig-out/bin/graff").resolve())