script-build-and-framework-tmp/agent/tests/test_coordinator_integration.py

import asyncio
from collections import deque

import pytest

from agent.orchestration.config import OrchestrationConfig
from agent.orchestration.coordinator import TaskConflict, TaskCoordinator
from agent.orchestration.models import (
    AgentRole,
    ArtifactRef,
    AttemptSubmission,
    CriterionResult,
    DecisionAction,
    TaskStatus,
    ValidationVerdict,
)
from agent.orchestration.protocols import ValidatorRunResult, WorkerRunResult
from agent.orchestration.store import FileSystemArtifactStore, FileSystemTaskStore, TraceEventSink
from agent.trace.goal_models import GoalTree
from agent.trace.models import Trace
from agent.trace.store import FileSystemTraceStore
from agent.core.runner import AgentRunner
from agent.orchestration.wiring import wire_orchestration


class FakeExecutor:
    def __init__(self, verdicts, submit_worker=True, submit_validator=True, delay=0):
        self.verdicts = deque(verdicts)
        self.submit_worker = submit_worker
        self.submit_validator = submit_validator
        self.delay = delay
        self.coordinator = None
        self.active = 0
        self.max_active = 0

    async def run_worker(self, context):
        self.active += 1
        self.max_active = max(self.max_active, self.active)
        if self.delay:
            await asyncio.sleep(self.delay)
        try:
            if self.submit_worker:
                await self.coordinator.submit_attempt(
                    {
                        **context,
                        "role": AgentRole.WORKER.value,
                        "trace_id": context["worker_trace_id"],
                        "tool_call_id": f"submit-{context['attempt_id']}",
                    },
                    AttemptSubmission(
                        summary="done",
                        artifact_refs=[ArtifactRef(uri=f"memory://{context['attempt_id']}", version="1")],
                    ),
                )
            return WorkerRunResult(context["worker_trace_id"], "completed")
        finally:
            self.active -= 1

    async def run_validator(self, context):
        verdict = self.verdicts.popleft() if self.verdicts else ValidationVerdict.PASSED
        if self.submit_validator:
            criteria = [
                CriterionResult(
                    criterion_id=item["criterion_id"], verdict=verdict,
                    reason="checked",
                )
                for item in context["task_spec"]["acceptance_criteria"]
            ]
            await self.coordinator.submit_validation(
                {
                    **context,
                    "role": AgentRole.VALIDATOR.value,
                    "trace_id": context["validator_trace_id"],
                    "tool_call_id": f"validate-{context['validation_id']}",
                },
                verdict,
                criteria,
                "independent report",
                [], [], [], "accept" if verdict == ValidationVerdict.PASSED else "replan",
            )
        return ValidatorRunResult(context["validator_trace_id"], "completed")


async def make_coordinator(tmp_path, executor):
    trace_store = FileSystemTraceStore(str(tmp_path))
    await trace_store.create_trace(Trace(trace_id="root", mode="agent", task="mission", agent_role="planner"))
    await trace_store.update_goal_tree("root", GoalTree(mission="mission"))
    task_store = FileSystemTaskStore(str(tmp_path))
    coordinator = TaskCoordinator(
        task_store,
        FileSystemArtifactStore(str(tmp_path)),
        trace_store,
        OrchestrationConfig(max_parallel_tasks=4),
        TraceEventSink(str(tmp_path)),
        executor,
    )
    executor.coordinator = coordinator
    await coordinator.ensure_ledger("root", "mission")
    return coordinator, task_store, trace_store


async def create_task(coordinator, objective="task", parent_task_id=None):
    result = await coordinator.create_tasks(
        "root",
        [{
            "objective": objective,
            "acceptance_criteria": [{"criterion_id": "c1", "description": "must pass", "hard": True}],
        }],
        parent_task_id=parent_task_id,
    )
    return result["tasks"][0]["task_id"]


@pytest.mark.asyncio
async def test_passed_validation_requires_planner_accept(tmp_path):
    executor = FakeExecutor([ValidationVerdict.PASSED])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    cycle = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    assert cycle.task_status == TaskStatus.AWAITING_DECISION
    ledger = await store.load("root")
    assert ledger.tasks[task_id].status == TaskStatus.AWAITING_DECISION

    await coordinator.decide_task(
        "root", task_id, cycle.validation_id, DecisionAction.ACCEPT,
        {"reason": "all hard criteria passed"}, "decision-1",
    )
    assert (await store.load("root")).tasks[task_id].status == TaskStatus.COMPLETED


@pytest.mark.asyncio
async def test_failed_or_inconclusive_cannot_be_accepted(tmp_path):
    executor = FakeExecutor([ValidationVerdict.FAILED])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    cycle = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    with pytest.raises(TaskConflict, match="passed"):
        await coordinator.decide_task(
            "root", task_id, cycle.validation_id, DecisionAction.ACCEPT,
            {"reason": "override"}, "bad-accept",
        )
    assert (await store.load("root")).tasks[task_id].status == TaskStatus.AWAITING_DECISION


@pytest.mark.asyncio
async def test_repair_is_limited_to_once_per_spec_version(tmp_path):
    executor = FakeExecutor([ValidationVerdict.FAILED, ValidationVerdict.FAILED])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    first = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    await coordinator.decide_task(
        "root", task_id, first.validation_id, DecisionAction.REPAIR,
        {"reason": "small local correction"}, "repair-1",
    )
    second = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    ledger = await store.load("root")
    attempts = [ledger.attempts[x] for x in ledger.tasks[task_id].attempt_ids]
    assert attempts[0].worker_trace_id == attempts[1].worker_trace_id
    with pytest.raises(TaskConflict, match="limit"):
        await coordinator.decide_task(
            "root", task_id, second.validation_id, DecisionAction.REPAIR,
            {"reason": "try again"}, "repair-2",
        )


@pytest.mark.asyncio
async def test_worker_without_submit_attempt_needs_replan(tmp_path):
    executor = FakeExecutor([], submit_worker=False)
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    result = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    ledger = await store.load("root")
    assert result.task_status == TaskStatus.NEEDS_REPLAN
    assert ledger.tasks[task_id].status == TaskStatus.NEEDS_REPLAN
    assert ledger.attempts[result.attempt_id].status.value == "failed"


@pytest.mark.asyncio
async def test_validator_without_submit_is_error_then_revalidates_with_new_trace(tmp_path):
    executor = FakeExecutor([ValidationVerdict.PASSED], submit_validator=False)
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    first = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    ledger = await store.load("root")
    first_validation = ledger.validations[first.validation_id]
    assert first.task_status == TaskStatus.NEEDS_REPLAN
    assert first_validation.status.value == "error"
    assert first_validation.verdict is None

    executor.submit_validator = True
    executor.verdicts.append(ValidationVerdict.PASSED)
    second = await coordinator.revalidate_attempt(
        "root", task_id, first.attempt_id, "revalidate-1"
    )
    assert second.task_status == TaskStatus.AWAITING_DECISION
    assert second.validation_id != first.validation_id
    assert second.validation.validator_trace_id != first_validation.validator_trace_id


@pytest.mark.asyncio
async def test_retry_uses_new_trace_and_revise_invalidates_old_validation(tmp_path):
    executor = FakeExecutor([
        ValidationVerdict.FAILED,
        ValidationVerdict.INCONCLUSIVE,
        ValidationVerdict.PASSED,
    ])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator)
    first = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    await coordinator.decide_task(
        "root", task_id, first.validation_id, DecisionAction.RETRY,
        {"reason": "use a fresh worker"}, "retry-1",
    )
    second = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    ledger = await store.load("root")
    first_attempt = ledger.attempts[first.attempt_id]
    second_attempt = ledger.attempts[second.attempt_id]
    assert first_attempt.worker_trace_id != second_attempt.worker_trace_id

    await coordinator.decide_task(
        "root", task_id, second.validation_id, DecisionAction.REVISE,
        {
            "reason": "clarify criterion",
            "objective": "revised task",
            "acceptance_criteria": [{"criterion_id": "c1", "description": "revised", "hard": True}],
        },
        "revise-1",
    )
    ledger = await store.load("root")
    assert ledger.tasks[task_id].current_spec_version == 2
    third = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    with pytest.raises(TaskConflict, match="obsolete"):
        await coordinator.decide_task(
            "root", task_id, first.validation_id, DecisionAction.ACCEPT,
            {"reason": "stale"}, "stale-accept",
        )
    await coordinator.decide_task(
        "root", task_id, third.validation_id, DecisionAction.ACCEPT,
        {"reason": "current validation passed"}, "current-accept",
    )


@pytest.mark.asyncio
async def test_block_unblock_cancel_and_supersede(tmp_path):
    executor = FakeExecutor([])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_id = await create_task(coordinator, "blocked")
    await coordinator.decide_task(
        "root", task_id, None, DecisionAction.BLOCK,
        {"reason": "external dependency"}, "block-1",
    )
    assert (await store.load("root")).tasks[task_id].blocked_reason == "external dependency"
    await coordinator.decide_task(
        "root", task_id, None, DecisionAction.UNBLOCK, {}, "unblock-1"
    )
    assert (await store.load("root")).tasks[task_id].status == TaskStatus.NEEDS_REPLAN
    await coordinator.decide_task(
        "root", task_id, None, DecisionAction.CANCEL,
        {"reason": "no longer needed"}, "cancel-1",
    )
    assert (await store.load("root")).tasks[task_id].status == TaskStatus.CANCELLED

    old_id = await create_task(coordinator, "old")
    result = await coordinator.decide_task(
        "root", old_id, None, DecisionAction.SUPERSEDE,
        {"reason": "replace spec", "replacement": {"objective": "replacement"}},
        "supersede-1",
    )
    replacement_id = result["payload"]["replacement_task_id"]
    ledger = await store.load("root")
    assert ledger.tasks[old_id].status == TaskStatus.SUPERSEDED
    assert ledger.tasks[replacement_id].status == TaskStatus.PENDING


@pytest.mark.asyncio
async def test_insert_after_keeps_stable_ids_and_updates_display_order(tmp_path):
    executor = FakeExecutor([])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    first = await create_task(coordinator, "first")
    third = await create_task(coordinator, "third")
    inserted = await coordinator.create_tasks(
        "root",
        [{"objective": "second"}],
        placement={"after_task_id": first},
        idempotency_key="insert",
    )
    second = inserted["tasks"][0]["task_id"]
    ledger = await store.load("root")
    assert ledger.tasks[first].display_path == "1"
    assert ledger.tasks[second].display_path == "2"
    assert ledger.tasks[third].display_path == "3"
    assert len({first, second, third}) == 3
    repeated = await coordinator.create_tasks(
        "root",
        [{"objective": "second"}],
        placement={"after_task_id": first},
        idempotency_key="insert",
    )
    assert repeated["tasks"][0]["task_id"] == second


@pytest.mark.asyncio
async def test_split_children_do_not_auto_complete_parent(tmp_path):
    executor = FakeExecutor([ValidationVerdict.FAILED, ValidationVerdict.PASSED, ValidationVerdict.PASSED])
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    parent_id = await create_task(coordinator, "parent")
    first = (await coordinator.dispatch_tasks("root", [parent_id]))[0]
    decision = await coordinator.decide_task(
        "root", parent_id, first.validation_id, DecisionAction.SPLIT,
        {
            "reason": "split work",
            "tasks": [
                {"objective": "child one", "acceptance_criteria": [{"criterion_id": "c1", "description": "pass"}]},
                {"objective": "child two", "acceptance_criteria": [{"criterion_id": "c1", "description": "pass"}]},
            ],
        },
        "split-1",
    )
    child_ids = decision["payload"]["child_task_ids"]
    cycles = await coordinator.dispatch_tasks("root", child_ids)
    for child_id, cycle in zip(child_ids, cycles):
        await coordinator.decide_task(
            "root", child_id, cycle.validation_id, DecisionAction.ACCEPT,
            {"reason": "passed"}, f"accept-{child_id}",
        )
    ledger = await store.load("root")
    assert all(ledger.tasks[x].status == TaskStatus.COMPLETED for x in child_ids)
    assert ledger.tasks[parent_id].status == TaskStatus.NEEDS_REPLAN


@pytest.mark.asyncio
async def test_parallel_tasks_are_isolated_and_bounded(tmp_path):
    executor = FakeExecutor([ValidationVerdict.PASSED] * 4, delay=0.02)
    coordinator, store, _ = await make_coordinator(tmp_path, executor)
    task_ids = [await create_task(coordinator, f"task-{i}") for i in range(4)]
    results = await coordinator.dispatch_tasks("root", task_ids, idempotency_key="batch")
    assert [x.task_id for x in results] == task_ids
    assert executor.max_active <= 4
    assert all(x.task_status == TaskStatus.AWAITING_DECISION for x in results)
    ledger = await store.load("root")
    assert len({ledger.attempts[ledger.tasks[x].attempt_ids[-1]].worker_trace_id for x in task_ids}) == 4

    repeated = await coordinator.dispatch_tasks("root", task_ids, idempotency_key="batch")
    assert [x.attempt_id for x in repeated] == [x.attempt_id for x in results]


@pytest.mark.asyncio
async def test_real_runner_local_executor_creates_independent_terminal_traces(tmp_path):
    import json

    async def fake_llm(messages, tools, **kwargs):
        names = {item["function"]["name"] for item in tools or []}
        if "submit_attempt" in names:
            arguments = {
                "summary": "implemented",
                "artifact_refs": [{"uri": "memory://result", "version": "1"}],
                "evidence_refs": [],
            }
            tool_name = "submit_attempt"
        elif "submit_validation" in names:
            arguments = {
                "verdict": "passed",
                "criterion_results": [{"criterion_id": "c1", "verdict": "passed", "reason": "verified"}],
                "summary": "independent pass",
                "evidence_refs": [],
                "unverified_claims": [],
                "risks": [],
                "recommendation": "accept",
            }
            tool_name = "submit_validation"
        else:
            raise AssertionError(f"unexpected tools: {names}")
        return {
            "content": "",
            "tool_calls": [{
                "id": f"call-{tool_name}",
                "type": "function",
                "function": {"name": tool_name, "arguments": json.dumps(arguments)},
            }],
            "finish_reason": "tool_calls",
        }

    trace_store = FileSystemTraceStore(str(tmp_path))
    await trace_store.create_trace(Trace(trace_id="root", mode="agent", task="mission", agent_role="planner"))
    await trace_store.update_goal_tree("root", GoalTree(mission="mission"))
    runner = AgentRunner(trace_store=trace_store, llm_call=fake_llm)
    coordinator = wire_orchestration(
        runner,
        FileSystemTaskStore(str(tmp_path)),
        FileSystemArtifactStore(str(tmp_path)),
    )
    await coordinator.ensure_ledger("root", "mission")
    task_id = await create_task(coordinator)
    cycle = (await coordinator.dispatch_tasks("root", [task_id]))[0]
    ledger = await coordinator.task_store.load("root")
    worker = await trace_store.get_trace(ledger.attempts[cycle.attempt_id].worker_trace_id)
    validator = await trace_store.get_trace(cycle.validation.validator_trace_id)
    assert cycle.task_status == TaskStatus.AWAITING_DECISION
    assert worker.agent_role == "worker" and worker.result_summary
    assert validator.agent_role == "validator" and validator.result_summary
    assert worker.trace_id != validator.trace_id