mirror of
https://github.com/razzant/ouroboros.git
synced 2026-10-03 12:18:39 +00:00
14fdc81343
Remove the universal commit-or-revert delivery guard in an explicit follow-up. Wake task waits on the actor mailbox without acknowledging messages, keep promoted lineage and observed execution facts, and retain full completion sources before synthesis. Cancellation observations remain separate from caller intent and never change stop authority. Co-authored-by: Ouroboros <311266734+ouroboros-agent@users.noreply.github.com>
110 lines
6.3 KiB
Python
110 lines
6.3 KiB
Python
"""Existing observation records carry their actual task, call and start facts."""
|
|
|
|
import json
|
|
import queue
|
|
from datetime import datetime, timedelta, timezone
|
|
from pathlib import Path
|
|
from types import SimpleNamespace
|
|
|
|
from ouroboros.agent import OuroborosAgent
|
|
from ouroboros.agent_task_pipeline import _store_task_result
|
|
from ouroboros.task_results import load_task_result, write_task_result
|
|
|
|
|
|
def test_known_start_survives_finalization_and_legacy_ts_is_not_relabelled(tmp_path):
|
|
env = SimpleNamespace(drive_root=tmp_path, repo_dir=tmp_path)
|
|
task = {"id": "known", "type": "task", "text": "work", "queued_at": 1700000000.0}
|
|
write_task_result(tmp_path, "known", "scheduled", ts="2020-01-01T00:00:00+00:00")
|
|
host = SimpleNamespace(env=env, _task_started_ts=1700000005.0)
|
|
OuroborosAgent._persist_running_record(host, task)
|
|
started = load_task_result(tmp_path, "known")["started_at"]
|
|
assert started == datetime.fromtimestamp(host._task_started_ts, timezone.utc).isoformat()
|
|
OuroborosAgent._persist_running_record(host, task) # preflight amendment is the same start
|
|
_store_task_result(env, task, "done", {"rounds": 1, "cost": 0}, {"tool_calls": []})
|
|
stored = load_task_result(tmp_path, "known")
|
|
assert stored["started_at"] == started and stored["queued_at"] == task["queued_at"]
|
|
assert stored["ts"] != started
|
|
write_task_result(tmp_path, "legacy", "running", ts="2020-01-01T00:00:00+00:00")
|
|
_store_task_result(env, {"id": "legacy", "type": "task"}, "done", {}, {"tool_calls": []})
|
|
assert "started_at" not in load_task_result(tmp_path, "legacy")
|
|
|
|
|
|
def test_real_llm_round_and_usage_join_across_worker_and_supervisor_logs(tmp_path):
|
|
from ouroboros.loop_llm_call import call_llm_with_retry
|
|
from supervisor.events_budget import _handle_llm_usage
|
|
|
|
worker, canonical = tmp_path / "worker", tmp_path / "canonical"
|
|
logs = worker / "logs"
|
|
logs.mkdir(parents=True)
|
|
events = queue.Queue()
|
|
class LLM:
|
|
def chat(self, **kwargs):
|
|
return ({"content": "done", "tool_calls": [], "finish_reason": "stop"},
|
|
{"provider": "openai-compatible", "resolved_model": "local-model", "cost": 0.0,
|
|
"prompt_tokens": 12, "completion_tokens": 3})
|
|
message, cost = call_llm_with_retry(LLM(), [{"role": "user", "content": "do work"}],
|
|
"openai-compatible::local-model", None, "medium", 1,
|
|
logs, "call-task", 7, events, {})
|
|
assert message["content"] == "done" and cost == 0
|
|
queued = list(events.queue)
|
|
usage_event = next(row for row in queued if row.get("type") == "llm_usage")
|
|
ctx = SimpleNamespace(DRIVE_ROOT=canonical, RUNNING={}, update_budget_from_usage=lambda usage: None,
|
|
bridge=SimpleNamespace(push_log=lambda event: None))
|
|
_handle_llm_usage(usage_event, ctx)
|
|
local = [json.loads(line) for line in (logs / "events.jsonl").read_text().splitlines()]
|
|
global_rows = [json.loads(line) for line in (canonical / "logs" / "events.jsonl").read_text().splitlines()]
|
|
round_row = next(row for row in local if row["type"] == "llm_round")
|
|
saved = next(row for row in global_rows if row["type"] == "llm_usage")
|
|
for key in ("llm_call_id", "execution_id", "round_id", "round"):
|
|
assert saved[key] == round_row[key]
|
|
assert saved["round"] == 7
|
|
assert saved["cost"] == 0 and saved["cost_known"] is True
|
|
assert not any(row["type"] == "llm_round" for row in global_rows)
|
|
|
|
|
|
def test_intrinsic_checkpoint_discloses_the_same_binding_ceiling_as_text(monkeypatch):
|
|
from ouroboros import task_pacing as pacing
|
|
|
|
monkeypatch.setattr(pacing, "get_pacing_interval_sec", lambda: 1)
|
|
ceiling = pacing.CostCeiling(state="active", ceiling_usd=80, root_cap_usd=100, basis="root_cap")
|
|
now = datetime.now(timezone.utc)
|
|
note = pacing.build_intrinsic_pacing_note(SimpleNamespace(_cost_ceiling=ceiling),
|
|
created=now - timedelta(seconds=60), now=now, round_idx=4, accumulated_usage={"cost": 10},
|
|
tree_cost_provider=lambda: {"accounted_usd": 20, "root_limit_usd": 100})
|
|
assert note.checkpoint["cost_ceiling"] == pacing.cost_ceiling_disclosure(ceiling)
|
|
assert note.checkpoint["tree_cap_usd"] == 100
|
|
assert "$80.00 in-task cost ceiling" in note.text
|
|
|
|
|
|
def test_terminal_delegation_rows_keep_existing_root_and_parent(tmp_path, monkeypatch):
|
|
from ouroboros import delegate_custody as custody
|
|
|
|
monkeypatch.setattr(custody, "_CUSTODY", {})
|
|
row = custody.RunCustody(run_id="correlated", task_id="child", root_task_id="root", parent_task_id="parent",
|
|
route_id="route", ledger_root=str(tmp_path))
|
|
custody.record_started(tmp_path, row)
|
|
custody.settle_run(tmp_path, SimpleNamespace(), row,
|
|
{"summary": {"state": "succeeded", "model": "test-model", "spendUsd": 0, "spendEstimated": False}})
|
|
custody.record_patch_disposed(tmp_path, row, disposition="rejected")
|
|
row.output_artifact, row.output_complete = "recorded-output.txt", True
|
|
assert custody.record_settled_unread(tmp_path, row)
|
|
records = [json.loads(line) for line in custody.event_log_path(tmp_path).read_text().splitlines()]
|
|
wanted = {custody.LEDGER_RECORDED, custody.SETTLED, custody.PATCH_DISPOSED, custody.SETTLED_UNREAD}
|
|
found = [record for record in records if record["type"] in wanted]
|
|
assert {record["type"] for record in found} == wanted
|
|
assert all(record["root_task_id"] == "root" and record["parent_task_id"] == "parent" for record in found)
|
|
|
|
|
|
def test_tool_error_manifest_has_typed_code_and_redacted_reason(tmp_path):
|
|
from ouroboros.loop_tool_execution import _execute_single_tool
|
|
from ouroboros.tools.tool_result import ToolResult
|
|
|
|
tools = SimpleNamespace(CODE_TOOLS=set(), _ctx=SimpleNamespace(task_metadata={}),
|
|
execute_result=lambda *_a: ToolResult(status="error", code="EXECUTOR_ERROR", text="Database initialization failed."))
|
|
logs = tmp_path / "logs"
|
|
logs.mkdir()
|
|
result = _execute_single_tool(tools, {"id": "tc1", "function": {"name": "probe", "arguments": "{}"}}, logs, "task")
|
|
manifest = json.loads(Path(result["trace_ref"]["manifest_ref"]["path"]).read_text())
|
|
assert manifest["tool_code"] == "EXECUTOR_ERROR"
|
|
assert manifest["error_preview"] == "Database initialization failed."
|
|
assert manifest["semantic_ok"] is False
|