multi-agent-mux/tests/test_issue5_phase2b_autocomplete.py

#!/usr/bin/env python3
"""Issue #5 Phase 2b — flag-gated auto-completion.

Characterize first, as in 2a: the interesting property here is what the system
*refuses* to do, so the refusals are pinned before anything is added that could
weaken them.

Sections
  1. X-01..X-05  the fail-closed preconditions as they stand today.
  2. finalize    the completion manifest: written by the agent, re-derived here.
  3. autocomplete the gate, and everything that must stop it.
  4. notifier    durable pending notification and restart recovery.
  5. flags       default off, evidence-backed allowlist, per-job opt-in.
"""
from __future__ import annotations

import hashlib
import json
import os
import subprocess
import sys
from pathlib import Path

import pytest

REPO_ROOT = Path(__file__).resolve().parent.parent
SKILLS = REPO_ROOT / ".agents" / "skills"
WATCHER_FIXTURES = REPO_ROOT / "tests" / "fixtures" / "issue_5" / "watcher"


def _reg(tmp_path) -> str:
    d = tmp_path / ".mam" / "jobs"
    d.mkdir(parents=True, exist_ok=True)
    return str(d)


def _make_job(registry_dir, **over):
    from lib_py.delegation import registry
    return registry.register_job(over.pop("prompt", "p"), registry_dir=registry_dir, **over)


def _load(registry_dir, job_id):
    from lib_py.delegation import mqtt_common
    return mqtt_common.load_job(job_id, registry_dir)


def _fingerprint(path: Path) -> dict:
    out = {}
    for p in sorted(Path(path).rglob("*")):
        if p.is_file():
            out[str(p.relative_to(path))] = hashlib.sha256(p.read_bytes()).hexdigest()
    return out


def _scenario(name="finalized_candidate"):
    return json.loads((WATCHER_FIXTURES / (name + ".json")).read_text(encoding="utf-8"))


@pytest.fixture(autouse=True)
def _default_flags(monkeypatch):
    """Auto-completion is off unless a test turns it on. That is the product
    default too, and every test that does not opt in is therefore also a check
    that the default really is off."""
    monkeypatch.delenv("MAM_AUTO_COMPLETE", raising=False)
    monkeypatch.delenv("MAM_AUTO_COMPLETE_BACKENDS", raising=False)
    monkeypatch.setenv("MAM_ENGINE_CAS", "1")


# ==============================================================================
# 1. X-01..X-05 — the fail-closed preconditions, before anything is added
# ==============================================================================

def test_x01_a_job_without_a_manifest_is_never_approved():
    """The property that makes unsupported agents safe without any new branch."""
    import copy
    from lib_py.loop import policy

    scenario = _scenario()
    assert policy.decide(scenario).decision == "candidate_approved"
    stripped = copy.deepcopy(scenario)
    for item in stripped["timeline"]:
        (item.get("payload") or {}).pop("report_manifest", None)
    stripped["filesystem"]["manifest_committed"] = False
    decision = policy.decide(stripped)
    assert decision.decision == "hold" and "manifest" in decision.reason


def test_x02_no_production_code_writes_a_completion_manifest_yet():
    """Pins the starting point: the protocol is entirely new in phase 2b."""
    hits = []
    for path in sorted((SKILLS / "lib_py").rglob("*.py")):
        if path.name in ("finalize.py", "autocomplete.py"):
            continue
        text = path.read_text(encoding="utf-8")
        if "report-manifest" in text:
            hits.append(path.name)
    assert not hits, "manifest writers already exist: {}".format(hits)


def test_x03_the_record_carries_iteration_not_an_attempt_id(mam_sandbox, tmp_path, monkeypatch):
    """Why auto-completion does not use the engine's expected_attempt.

    The engine compares expected_attempt against `attempt_id` falling back to
    `iteration`. A record now carries attempt_id (initialized to 1 in Issue 6)
    and iteration, and the attempt guard is applied where attempt identity
    actually lives: the dispatch record in the shadow store, compared against
    observed evidence."""
    monkeypatch.chdir(mam_sandbox)
    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    record = _load(reg, job_id)
    assert "iteration" in record
    assert record.get("attempt_id") == 1


def test_x04_policy_rejects_evidence_from_a_superseded_attempt():
    """The attempt guard that auto-completion actually relies on."""
    import copy
    from lib_py.loop import policy

    scenario = _scenario()
    superseded = copy.deepcopy(scenario)
    superseded["dispatch_context"]["attempt_id"] = 2   # evidence still says 1
    decision = policy.decide(superseded)
    assert decision.decision == "hold" and "attempt_id" in decision.reason


def test_x05_the_explicit_event_path_is_untouched_by_default(mam_sandbox, tmp_path, monkeypatch):
    """An explicitly completed job looks exactly as it always did."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import mqtt_common

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    mqtt_common.update_job_status(job_id, reg, _source="publish_event", status="running")
    mqtt_common.update_job_status(job_id, reg, _source="publish_event", status="completed")
    record = _load(reg, job_id)
    assert record["status"] == "completed"
    assert "completed_by" not in record, "an explicit completion carries no auto marker"


# ==============================================================================
# 2. finalize — declared by the agent, re-derived by the reader
# ==============================================================================

def _report(tmp_path, body="done\n", staged=False):
    d = tmp_path / "claude-reports"
    d.mkdir(parents=True, exist_ok=True)
    path = d / "report-final.md"
    (path.with_name(path.name + ".tmp") if staged else path).write_text(body, encoding="utf-8")
    return path


def test_finalize_promotes_a_staged_report_atomically(tmp_path):
    """A reader must never see a half-written report as finalized."""
    from lib_py.delegation import finalize

    path = _report(tmp_path, "final body\n", staged=True)
    assert not path.exists() and path.with_name(path.name + ".tmp").exists()
    manifest = finalize.finalize_report(path, "job1", 1)
    assert path.exists() and not path.with_name(path.name + ".tmp").exists()
    assert manifest["sha256"] == finalize.sha256_of(path)


def test_finalize_verifies_against_the_file_not_the_declaration(tmp_path):
    """The hash is recomputed, so a forged manifest cannot certify itself."""
    from lib_py.delegation import finalize

    path = _report(tmp_path)
    finalize.finalize_report(path, "job1", 1)
    assert finalize.verify_manifest(path.parent, "job1").ok

    manifest_file = finalize.manifest_in(path.parent)
    forged = json.loads(manifest_file.read_text(encoding="utf-8"))
    forged["sha256"] = "0" * 64
    manifest_file.write_text(json.dumps(forged), encoding="utf-8")
    verdict = finalize.verify_manifest(path.parent, "job1")
    assert not verdict.ok and "changed after finalization" in verdict.reason


def test_finalize_is_invalidated_by_a_rewritten_report(tmp_path):
    """A new body needs a new finalization; the old hash is void immediately."""
    from lib_py.delegation import finalize

    path = _report(tmp_path, "first\n")
    finalize.finalize_report(path, "job1", 1)
    assert finalize.verify_manifest(path.parent, "job1").ok
    path.write_text("rewritten\n", encoding="utf-8")
    verdict = finalize.verify_manifest(path.parent, "job1")
    assert not verdict.ok and "changed after finalization" in verdict.reason


@pytest.mark.parametrize("kwargs,fragment", [
    ({"job_id": "other"}, "belongs to job"),
    ({"expected_attempt": 9}, "belongs to attempt"),
])
def test_finalize_rejects_a_manifest_for_a_different_job_or_attempt(tmp_path, kwargs, fragment):
    from lib_py.delegation import finalize
    path = _report(tmp_path)
    finalize.finalize_report(path, "job1", 1)
    args = {"job_id": "job1"}
    args.update(kwargs)
    verdict = finalize.verify_manifest(path.parent, **args)
    assert not verdict.ok and fragment in verdict.reason


def test_finalize_requires_evidence_for_every_declared_side_effect(tmp_path):
    """Partial evidence is no evidence: the work is not known to be complete."""
    from lib_py.delegation import finalize

    path = _report(tmp_path)
    finalize.finalize_report(path, "job1", 1,
                             side_effects=finalize.parse_side_effects(["git_push=abc123"]))
    ok = finalize.verify_manifest(path.parent, "job1", expected_artifacts=["git_push"])
    assert ok.ok
    missing = finalize.verify_manifest(path.parent, "job1",
                                       expected_artifacts=["git_push", "api_call"])
    assert not missing.ok and "api_call" in missing.reason


def test_finalize_rejects_a_malformed_side_effect_pair():
    from lib_py.delegation import finalize
    with pytest.raises(ValueError):
        finalize.parse_side_effects(["nokey"])
    with pytest.raises(ValueError):
        finalize.parse_side_effects(["kind="])


def test_finalize_missing_manifest_is_a_refusal_not_a_crash(tmp_path):
    from lib_py.delegation import finalize
    d = tmp_path / "empty"
    d.mkdir()
    verdict = finalize.verify_manifest(d, "job1")
    assert not verdict.ok and "no completion manifest" in verdict.reason


def test_publish_event_finalizes_before_it_publishes(mam_sandbox, tmp_path, monkeypatch):
    """The CLI surface an agent actually calls."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import publish_event
    from lib_py.delegation import finalize

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    record = _load(reg, job_id)
    record["broker"] = {"host": "127.0.0.1", "port": 65432, "tls": False}
    Path(reg, job_id + ".json").write_text(json.dumps(record), encoding="utf-8")

    path = _report(tmp_path, "the report\n", staged=True)
    rc = publish_event.main([
        "--registry-dir", reg, "--job", job_id, "--event", "completed",
        "--attempts", "1", "--finalize-report", str(path),
        "--side-effect", "git_push=deadbeef",
    ])
    assert rc == 2, "the broker is unreachable, but finalization still happened"
    assert path.exists(), "the staged report was promoted"
    verdict = finalize.verify_manifest(path.parent, job_id)
    assert verdict.ok
    assert verdict.manifest["side_effects"][0]["kind"] == "git_push"


def test_publish_event_reports_a_bad_side_effect_instead_of_publishing(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import publish_event
    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    path = _report(tmp_path)
    rc = publish_event.main([
        "--registry-dir", reg, "--job", job_id, "--event", "completed",
        "--finalize-report", str(path), "--side-effect", "malformed",
    ])
    assert rc == 1


# ==============================================================================
# 3. autocomplete — the gate, and everything that must stop it
# ==============================================================================

def _ready_job(tmp_path, mam_sandbox, *, backend="claude-code", opt_in=True,
               artifacts=None, body="report\n"):
    """A job in exactly the state that should auto-complete, plus its evidence."""
    from lib_py.delegation import finalize
    from lib_py.loop import store

    reg = _reg(tmp_path)
    job_id = _make_job(reg, agent=backend, expected_artifacts=artifacts or [])
    from lib_py.delegation import engine
    engine.transition(job_id, "claim", registry_dir=reg)
    if opt_in:
        engine.transition(job_id, "progress", registry_dir=reg,
                          fields={"auto_complete": True})

    scenario = _scenario()
    dispatch = scenario["dispatch_context"]
    dispatch["job_id"] = job_id
    for item in scenario["timeline"]:
        env = item.get("collector_attribution")
        if env:
            env["job_id"] = job_id
        # A non-herdr payload is a legitimate identifier source, so it has to
        # agree too. Leaving the fixture's original id here is a contradiction,
        # and the policy correctly refuses it.
        payload = item.get("payload") or {}
        if item.get("source_kind") != "herdr" and "job_id" in payload:
            payload["job_id"] = job_id
    # The dispatch record carries the record's own token for this attempt, which
    # is what ties the evidence to the attempt that is currently running.
    from lib_py.delegation import engine as _engine
    token = _load(reg, job_id).get(_engine.DISPATCH_TOKEN_FIELD)
    store.record_dispatch(job_id, dispatch["attempt_id"], dispatch["agent_session_uuid"],
                          dispatch["pane_process_generation"], workspace=str(mam_sandbox),
                          dispatch_token=token)
    store.mark_execution_anchor(job_id, workspace=str(mam_sandbox))

    report = _report(tmp_path, body)
    finalize.finalize_report(report, job_id, dispatch["attempt_id"],
                             side_effects=finalize.parse_side_effects(
                                 ["{}=ref".format(a) for a in (artifacts or [])]))
    return job_id, reg, scenario, report.parent


def test_gate_commits_when_every_condition_holds(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed, result.reason
    record = _load(reg, job_id)
    assert record["status"] == "completed"
    assert record["completed_by"] == "auto_complete", "an inferred terminal must be marked"
    assert record["pending_notification"]["event"] == "completed"


def test_g4_default_off_commits_nothing_and_leaves_the_registry_untouched(
    mam_sandbox, tmp_path, monkeypatch
):
    """G4: with the flag unset, the registry must not change at all."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    before = _fingerprint(Path(reg))
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False
    assert "disabled" in result.reason
    assert _fingerprint(Path(reg)) == before, "the default must write nothing"


@pytest.mark.parametrize("backend", ["agy", "hermes", "opencode", "unknown-agent"])
def test_g9_a_backend_without_quiescence_evidence_never_auto_completes(
    mam_sandbox, tmp_path, monkeypatch, backend
):
    """Inferring 'finished' from stillness needs stillness to have been observed."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox, backend=backend)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "quiescent-state evidence" in result.reason
    assert _load(reg, job_id)["status"] == "running"


def test_the_allowlist_can_be_narrowed_but_never_widened(monkeypatch):
    from lib_py.loop import autocomplete

    monkeypatch.setenv("MAM_AUTO_COMPLETE_BACKENDS", "claude")
    assert autocomplete.allowed_backends() == ("claude",)
    monkeypatch.setenv("MAM_AUTO_COMPLETE_BACKENDS", "claude,hermes,opencode")
    assert autocomplete.allowed_backends() == ("claude",), "unobserved backends are dropped"


def test_a_job_that_did_not_opt_in_is_not_auto_completed(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox, opt_in=False)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "opt in" in result.reason


def test_g5_a_rewritten_report_stops_the_gate(mam_sandbox, tmp_path, monkeypatch):
    """G5: the manifest is checked against the file, so a late edit disqualifies."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    (report_dir / "report-final.md").write_text("rewritten after finalization\n", encoding="utf-8")
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "changed after finalization" in result.reason
    assert _load(reg, job_id)["status"] == "running"


def test_g5_missing_side_effect_evidence_stops_the_gate(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete
    from lib_py.delegation import finalize

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox,
                                                   artifacts=["git_push"])
    # re-declare without the evidence the job said it would produce
    finalize.finalize_report(report_dir / "report-final.md", job_id, 1, side_effects=[])
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "git_push" in result.reason


def test_d7_a_policy_that_cannot_decide_holds(mam_sandbox, tmp_path, monkeypatch):
    """'Could not decide' is never 'done'."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete, policy

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)

    def explode(_scenario):
        raise RuntimeError("evidence is contradictory")

    monkeypatch.setattr(policy, "decide", explode)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "policy error" in result.reason
    assert _load(reg, job_id)["status"] == "running"


def test_a_late_explicit_terminal_beats_the_gate(mam_sandbox, tmp_path, monkeypatch):
    """If the agent already reported, the boundary refuses and that is correct."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.delegation import mqtt_common
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    mqtt_common.update_job_status(job_id, reg, _source="publish_event", status="error")
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False
    record = _load(reg, job_id)
    assert record["status"] == "error", "the explicit report wins"
    assert "completed_by" not in record


# ==============================================================================
# 4. notifier — the intent is durable, and recovery re-sends it
# ==============================================================================

def test_the_notification_intent_is_committed_with_the_state(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")
    record = _load(reg, job_id)
    assert record["status"] == "completed"
    intent = notifier.pending(job_id, reg)
    assert intent and intent["event"] == "completed" and intent["seq"] == record["transition_seq"]


def test_flush_publishes_then_settles(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")

    sent = []
    assert notifier.flush(job_id, reg, publish=lambda j, i, r: sent.append(i) or 0)
    assert len(sent) == 1 and sent[0]["event"] == "completed"
    assert notifier.pending(job_id, reg) is None
    # settling is idempotent: a second flush has nothing to do
    assert notifier.flush(job_id, reg, publish=lambda *a: 1 / 0)


def test_a_failed_send_leaves_the_intent_pending(mam_sandbox, tmp_path, monkeypatch):
    """An unsent notification that looks sent is worse than one that is retried."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")

    assert notifier.flush(job_id, reg, publish=lambda *a: 2) is False
    assert notifier.pending(job_id, reg) is not None
    assert notifier.flush(job_id, reg, publish=lambda *a: (_ for _ in ()).throw(OSError())) is False
    assert notifier.pending(job_id, reg) is not None
    assert _load(reg, job_id)["status"] == "completed", "the state stays committed regardless"


def test_g7_restart_recovery_resends_every_owed_notification(mam_sandbox, tmp_path, monkeypatch):
    """G7: the crash window between committing and publishing."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    owed = []
    for _ in range(3):
        job_id = _make_job(reg)
        engine.transition(job_id, "claim", registry_dir=reg)
        engine.transition(job_id, "completed", registry_dir=reg, notify="completed")
        owed.append(job_id)
    settled = _make_job(reg)
    engine.transition(settled, "claim", registry_dir=reg)

    assert sorted(notifier.jobs_with_pending(reg)) == sorted(owed)

    sent = []
    results = notifier.recover(reg, publish=lambda j, i, r: sent.append(j) or 0)
    assert all(results.values()) and sorted(sent) == sorted(owed)
    assert notifier.jobs_with_pending(reg) == [], "nothing is owed after recovery"


# ==============================================================================
# 6. Cross-review regressions (job f63d8279)
#
# Three reviewers, seven blocking items, all reproduced before being accepted.
# The theme: a gate whose strictness depended on the caller, and a boundary that
# trusted its only current caller instead of enforcing its own rule.
# ==============================================================================

def test_f1_an_omitted_report_directory_is_a_refusal_not_a_shortcut(
    mam_sandbox, tmp_path, monkeypatch
):
    """Independent verification cannot be optional: skipping it puts the
    collector's own flags back in charge of the decision."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, _report_dir = _ready_job(tmp_path, mam_sandbox)
    before = _fingerprint(Path(reg))
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox))   # no report_dir
    assert result.committed is False and "independently verify" in result.reason
    assert _fingerprint(Path(reg)) == before, "a refusal must not write"


def test_f1_an_empty_report_directory_is_also_a_refusal(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, _rd = _ready_job(tmp_path, mam_sandbox)
    empty = tmp_path / "empty-reports"
    empty.mkdir()
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=empty)
    assert result.committed is False and "no completion manifest" in result.reason


def test_r2_evidence_for_another_job_cannot_complete_this_one(
    mam_sandbox, tmp_path, monkeypatch
):
    """The policy checks a scenario against the dispatch context inside that same
    scenario, which is self-consistent by construction. The gate must bind the
    evidence to the target."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    scenario["dispatch_context"]["job_id"] = "some-other-job"
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "not" in result.reason
    assert _load(reg, job_id)["status"] == "running"


@pytest.mark.parametrize("field", ["attempt_id", "agent_session_uuid", "pane_process_generation"])
def test_r2_evidence_must_match_the_recorded_dispatch(mam_sandbox, tmp_path, monkeypatch, field):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    scenario["dispatch_context"][field] = "not-what-was-dispatched"
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and field in result.reason


def test_r2_a_job_with_no_recorded_dispatch_is_refused(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    (Path(mam_sandbox) / ".mam" / "shadow" / "attribution" / (job_id + ".json")).unlink()
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "no recorded dispatch" in result.reason


def test_the_cas_rollback_switch_also_disables_auto_completion(mam_sandbox, tmp_path, monkeypatch):
    """Inferred completion is built on the CAS boundary, so committing while
    that boundary is rolled back would bypass it."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    monkeypatch.setenv("MAM_ENGINE_CAS", "0")
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "rolled back" in result.reason
    assert _load(reg, job_id)["status"] == "running"


@pytest.mark.parametrize("agent", ["claude-experimental", "codexish", "grok2"])
def test_o4_a_backend_key_must_match_exactly_not_by_prefix(
    mam_sandbox, tmp_path, monkeypatch, agent
):
    """A future name must not inherit evidence that was never collected for it."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox, backend=agent)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "quiescent-state evidence" in result.reason


def test_a_reviewer_without_a_verdict_token_is_not_completed(mam_sandbox, tmp_path, monkeypatch):
    """Phase 0 safety contract: a reviewer that has not written its verdict has
    not finished reviewing, and run_loop reads a missing token as NOT PASS."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.delegation import finalize
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(
        tmp_path, mam_sandbox, body="a thorough review with no verdict line\n")
    from lib_py.delegation import engine
    engine.transition(job_id, "progress", registry_dir=reg, fields={"role": "Reviewer"})

    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "verdict token" in result.reason
    assert _load(reg, job_id)["status"] == "running"

    # writing the verdict requires a fresh finalization, and then it passes
    report = report_dir / "report-final.md"
    report.write_text("a thorough review\n\n[VERDICT: PASS]\n", encoding="utf-8")
    finalize.finalize_report(report, job_id, scenario["dispatch_context"]["attempt_id"])
    ok = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                               workspace=str(mam_sandbox), report_dir=report_dir)
    assert ok.committed, ok.reason


def test_a_worker_role_needs_no_verdict_token(mam_sandbox, tmp_path, monkeypatch):
    """The token requirement is a reviewer contract, not a general one."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox,
                                                   body="worker output, no verdict\n")
    assert _load(reg, job_id)["role"] == "Worker"
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed, result.reason


def test_grok_f2_bookkeeping_cannot_merge_arbitrary_fields(mam_sandbox, tmp_path, monkeypatch):
    """A status-free request must not become an unrestricted merge on a record
    nothing else is allowed to touch."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")
    before = _load(reg, job_id)

    result = engine.transition(job_id, "notification_settled", registry_dir=reg,
                               fields={"pending_notification": None, "prompt": "HACKED",
                                       "agent_session": "herdr:attacker"},
                               source="not-an-operator")
    after = _load(reg, job_id)
    assert result.committed is False and "may not merge" in result.reason
    assert after["prompt"] == before["prompt"]
    assert after["agent_session"] == before["agent_session"]
    assert after["pending_notification"], "the intent survives a refused settle"

    # the legitimate settle still works
    from lib_py.delegation import notifier
    assert notifier.settle(job_id, reg, sent_intent=after["pending_notification"])
    assert notifier.pending(job_id, reg) is None


def test_r4_a_newer_intent_is_not_erased_by_a_slow_send(mam_sandbox, tmp_path, monkeypatch):
    """Publishing happens outside the lock, so a replacement can land mid-flight.
    Settling unconditionally would drop a notification nobody sent."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")

    def racing_publisher(job, intent, registry_dir):
        engine.transition(job, "notification_settled", registry_dir=registry_dir,
                          fields={"pending_notification":
                                  notifier.intent_for("error", 999, "later")})
        return 0

    notifier.flush(job_id, reg, publish=racing_publisher)
    still_owed = notifier.pending(job_id, reg)
    assert still_owed and still_owed["seq"] == 999, "the replacement must survive"


def test_r4_the_published_payload_carries_the_intent_identity(mam_sandbox, tmp_path, monkeypatch):
    """The transport sequence cannot identify a retry of the same intent, so the
    intent's own identity travels in the payload for consumers to dedupe on."""
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    record = _load(reg, job_id)
    record["broker"] = {"host": "127.0.0.1", "port": 65432, "tls": False}
    Path(reg, job_id + ".json").write_text(json.dumps(record), encoding="utf-8")
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")

    intent = notifier.pending(job_id, reg)
    notifier.flush(job_id, reg, attempts=1)   # real path; broker is unreachable
    from lib_py.delegation import mqtt_common
    events = list(mqtt_common.iter_logged_events(job_id, mqtt_common.get_logs_dir()))
    published = [e for e in events if e.get("event") == "published"]
    assert published, "the attempt is recorded even though the broker refused"
    carried = published[-1]["payload"]["data"].get("notification_intent")
    assert carried and carried["seq"] == intent["seq"]


@pytest.mark.parametrize("entry", [
    {"kind": "deploy"},
    {"kind": "deploy", "reference": "   "},
    {"reference": "abc"},
    "not-an-object",
])
def test_r3_a_side_effect_entry_must_carry_real_evidence(tmp_path, entry):
    """A kind with no reference names something without evidencing it."""
    from lib_py.delegation import finalize

    path = _report(tmp_path)
    with pytest.raises(ValueError):
        finalize.finalize_report(path, "job1", 1, side_effects=[entry])

    # and a hand-written manifest is rejected at verification too
    finalize.finalize_report(path, "job1", 1,
                             side_effects=[{"kind": "deploy", "reference": "sha"}])
    manifest_file = finalize.manifest_in(path.parent)
    forged = json.loads(manifest_file.read_text(encoding="utf-8"))
    forged["side_effects"] = [entry]
    manifest_file.write_text(json.dumps(forged), encoding="utf-8")
    verdict = finalize.verify_manifest(path.parent, "job1", expected_artifacts=["deploy"])
    assert not verdict.ok


def test_o2_the_manifest_records_an_absolute_report_path(tmp_path, monkeypatch):
    """A relative path would resolve against whatever directory the reader is in."""
    from lib_py.delegation import finalize

    path = _report(tmp_path)
    monkeypatch.chdir(tmp_path)
    manifest = finalize.finalize_report(Path("claude-reports/report-final.md"), "job1", 1)
    assert Path(manifest["report_path"]).is_absolute()
    monkeypatch.chdir(REPO_ROOT)
    assert finalize.verify_manifest(path.parent, "job1").ok, "verification survives a chdir"


# ==============================================================================
# 7. Cross-review round 2 regressions (job 4cfc1649)
#
# R2 (codex): attribution was checked before the terminal commit but nothing
#             bound the two, so a job could complete, be re-dispatched and
#             claimed again in between, and stale evidence finished the new
#             attempt.
# R4 (codex): the intent identity was emitted but the real subscriber ignored
#             it, so a replay with a fresh wire sequence was accepted twice.
# ==============================================================================

def test_r2_a_redispatch_between_the_check_and_the_commit_is_refused(
    mam_sandbox, tmp_path, monkeypatch
):
    """The exact interleaving: complete, re-dispatch, claim again, then let the
    original commit proceed. Evidence for attempt 1 must not finish attempt 2."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.delegation import engine
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)

    real_transition = engine.transition
    fired = {}

    def racing(job, request="merge", **kwargs):
        if request == "completed" and kwargs.get("source") == "auto_complete" and not fired:
            fired["yes"] = True
            engine.transition = real_transition
            real_transition(job, "completed", registry_dir=reg)
            real_transition(job, "redispatch", registry_dir=reg,
                            source="registry_cli", fields={"iteration": 2})
            real_transition(job, "claim", registry_dir=reg)
            engine.transition = racing
        return real_transition(job, request, **kwargs)

    monkeypatch.setattr(engine, "transition", racing)
    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    monkeypatch.setattr(engine, "transition", real_transition)

    record = _load(reg, job_id)
    assert result.committed is False and "dispatch token" in result.reason
    assert record["status"] == "running", "the new attempt must still be running"
    assert record["iteration"] == 2
    assert "completed_by" not in record


def test_r2_evidence_from_a_superseded_dispatch_is_refused(mam_sandbox, tmp_path, monkeypatch):
    """Even without a race: evidence recorded against an older token is stale."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.delegation import engine
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    # a full re-dispatch mints a new token; the recorded evidence keeps the old
    engine.transition(job_id, "completed", registry_dir=reg)
    engine.transition(job_id, "redispatch", registry_dir=reg,
                      source="registry_cli", fields={"iteration": 2})
    engine.transition(job_id, "claim", registry_dir=reg)

    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "superseded dispatch" in result.reason


def test_r2_a_job_without_a_dispatch_token_is_refused(mam_sandbox, tmp_path, monkeypatch):
    """Nothing to bind the decision to is a refusal, not a pass."""
    monkeypatch.chdir(mam_sandbox)
    monkeypatch.setenv("MAM_AUTO_COMPLETE", "1")
    from lib_py.delegation import engine, mqtt_common
    from lib_py.loop import autocomplete

    job_id, reg, scenario, report_dir = _ready_job(tmp_path, mam_sandbox)
    record = _load(reg, job_id)
    record.pop(engine.DISPATCH_TOKEN_FIELD, None)
    mqtt_common._atomic_write_record(job_id, reg, record)

    result = autocomplete.evaluate(job_id, scenario, registry_dir=reg,
                                   workspace=str(mam_sandbox), report_dir=report_dir)
    assert result.committed is False and "no dispatch token" in result.reason


def test_r2_a_new_attempt_mints_a_new_token(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine
    reg = _reg(tmp_path)
    job_id = _make_job(reg)

    engine.transition(job_id, "claim", registry_dir=reg)
    first = _load(reg, job_id)[engine.DISPATCH_TOKEN_FIELD]
    engine.transition(job_id, "progress", registry_dir=reg, fields={"note": "x"})
    assert _load(reg, job_id)[engine.DISPATCH_TOKEN_FIELD] == first, (
        "an ordinary transition must not mint a new attempt"
    )
    engine.transition(job_id, "completed", registry_dir=reg)
    engine.transition(job_id, "redispatch", registry_dir=reg,
                      source="registry_cli", fields={"iteration": 2})
    assert _load(reg, job_id)[engine.DISPATCH_TOKEN_FIELD] != first


def test_r2_the_token_precondition_is_enforced_by_the_engine(mam_sandbox, tmp_path, monkeypatch):
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine
    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    engine.transition(job_id, "claim", registry_dir=reg)
    before = _fingerprint(Path(reg))

    stale = engine.transition(job_id, "completed", registry_dir=reg,
                              expected_dispatch_token="not-the-current-token")
    assert stale.committed is False and "dispatch token changed" in stale.reason
    assert _fingerprint(Path(reg)) == before, "a refused precondition must not write"

    token = _load(reg, job_id)[engine.DISPATCH_TOKEN_FIELD]
    assert engine.transition(job_id, "completed", registry_dir=reg,
                             expected_dispatch_token=token).committed


def test_r4_the_subscriber_drops_a_replayed_notification_intent():
    """A replay is re-published with a fresh wire sequence, so the monotonic
    guard cannot see it; the stable identity is what must be deduplicated."""
    import types
    from lib_py.delegation import job_subscriber

    watcher = job_subscriber._Watcher({"j1"}, {"j1": None}, {"j1": 0})

    def deliver(wire_seq, intent_seq):
        payload = {"schema_version": 1, "seq": wire_seq, "job_id": "j1",
                   "event": "completed", "timestamp": "t", "detail": "d",
                   "data": {"notification_intent": {"seq": intent_seq,
                                                    "event": "completed",
                                                    "timestamp": "T"}}}
        watcher.on_message(None, None, types.SimpleNamespace(
            topic="python/mqtt/jobs/j1/events",
            payload=json.dumps(payload).encode()))

    deliver(1, 7)
    deliver(2, 7)          # the same intent, retried
    assert watcher.events.qsize() == 1, "a replayed intent must not be acted on twice"
    deliver(3, 8)          # a genuinely different notification
    assert watcher.events.qsize() == 2


def test_r4_events_without_an_intent_are_unaffected():
    """The existing sequence guard remains the rule for ordinary events."""
    import types
    from lib_py.delegation import job_subscriber

    watcher = job_subscriber._Watcher({"j1"}, {"j1": None}, {"j1": 0})

    def deliver(wire_seq):
        payload = {"schema_version": 1, "seq": wire_seq, "job_id": "j1",
                   "event": "progress", "timestamp": "t", "detail": "d", "data": {}}
        watcher.on_message(None, None, types.SimpleNamespace(
            topic="python/mqtt/jobs/j1/events",
            payload=json.dumps(payload).encode()))

    deliver(1)
    deliver(2)
    assert watcher.events.qsize() == 2
    deliver(2)             # non-increasing: still dropped by the sequence guard
    assert watcher.events.qsize() == 2


_MISSING = object()


def _intent_watcher(last_seq=0):
    from lib_py.delegation import job_subscriber
    return job_subscriber._Watcher({"j1"}, {"j1": None}, {"j1": last_seq})


def _deliver_intent(watcher, wire_seq, intent, event="completed"):
    """Hand one schema-v1 event to the real callback. `intent` is placed in
    data.notification_intent verbatim, so a malformed one can be modelled."""
    import types
    payload = {"schema_version": 1, "seq": wire_seq, "job_id": "j1",
               "event": event, "timestamp": "t", "detail": "d", "data": {}}
    if intent is not _MISSING:
        payload["data"]["notification_intent"] = intent
    watcher.on_message(None, None, types.SimpleNamespace(
        topic="python/mqtt/jobs/j1/events",
        payload=json.dumps(payload).encode()))


def test_r4b_a_stale_packet_does_not_burn_the_intent_for_its_own_retry():
    """The defect: the identity was recorded before the sequence guard ran, so a
    late packet marked its intent as seen and the legitimate retry that followed
    was mistaken for a replay of an event the subscriber never delivered."""
    watcher = _intent_watcher(last_seq=10)
    intent = {"seq": 7, "event": "completed", "timestamp": "T"}

    _deliver_intent(watcher, 9, intent)     # late/out-of-order: refused by the seq guard
    assert watcher.events.qsize() == 0
    assert watcher.seen_intents.get("j1", set()) == set(), (
        "a refused event must not leave its intent marked as accepted"
    )

    _deliver_intent(watcher, 11, intent)    # the genuine retry of that notification
    assert watcher.events.qsize() == 1, "the retry of a never-delivered intent must be accepted"

    _deliver_intent(watcher, 12, intent)    # now it really is a replay
    assert watcher.events.qsize() == 1


def test_r4b_a_duplicate_intent_does_not_advance_the_sequence_floor():
    """Dropping a replay must not consume sequence numbers: the wire seq of an
    event that was never delivered has no claim on the monotonic floor."""
    watcher = _intent_watcher()
    intent = {"seq": 7, "event": "completed", "timestamp": "T"}

    _deliver_intent(watcher, 1, intent)
    assert watcher.events.qsize() == 1 and watcher.last_seq["j1"] == 1
    _deliver_intent(watcher, 9, intent)     # replay, refused
    assert watcher.events.qsize() == 1
    assert watcher.last_seq["j1"] == 1, "a refused replay must not raise the sequence floor"

    _deliver_intent(watcher, 2, {"seq": 8, "event": "completed", "timestamp": "T"})
    assert watcher.events.qsize() == 2, "an ordinary event after a refused replay still lands"


def test_r4b_an_unhashable_intent_is_refused_instead_of_raising():
    """A callback that raises takes down the network loop for every watched job,
    so a malformed remote payload must be refused inside defensive parsing."""
    watcher = _intent_watcher()
    _deliver_intent(watcher, 1, {"seq": [1, 2], "event": "completed", "timestamp": "T"})
    assert watcher.events.qsize() == 0
    assert watcher.last_seq["j1"] == 0, "a refused event must not advance the floor"

    # A well-formed intent on the same job is still accepted afterwards.
    _deliver_intent(watcher, 2, {"seq": 7, "event": "completed", "timestamp": "T"})
    assert watcher.events.qsize() == 1


def test_r4b_a_non_object_intent_is_refused():
    """Present but not an object: no stable identity can be derived, so the
    replay guard could not protect it."""
    watcher = _intent_watcher()
    _deliver_intent(watcher, 1, "completed:7")
    assert watcher.events.qsize() == 0
    _deliver_intent(watcher, 2, ["completed", 7])
    assert watcher.events.qsize() == 0


def test_r4b_an_absent_or_null_intent_is_treated_as_no_intent():
    """The ordinary path is unchanged: events that carry no intent are governed
    by the sequence guard alone."""
    watcher = _intent_watcher()
    _deliver_intent(watcher, 1, _MISSING, event="progress")
    _deliver_intent(watcher, 2, None, event="progress")
    assert watcher.events.qsize() == 2
    assert watcher.seen_intents.get("j1", set()) == set()
    _deliver_intent(watcher, 2, None, event="progress")   # non-increasing
    assert watcher.events.qsize() == 2


def test_r4b_a_real_retry_is_deduplicated_by_the_real_consumer(mam_sandbox, tmp_path, monkeypatch):
    """End to end, no hand-built payload: the notifier's own retry of an unsent
    notification is handed to the real subscriber callback and must land once."""
    import types
    monkeypatch.chdir(mam_sandbox)
    from lib_py.delegation import engine, job_subscriber, mqtt_common, notifier

    reg = _reg(tmp_path)
    job_id = _make_job(reg)
    record = _load(reg, job_id)
    record["broker"] = {"host": "127.0.0.1", "port": 65432, "tls": False}
    Path(reg, job_id + ".json").write_text(json.dumps(record), encoding="utf-8")
    engine.transition(job_id, "claim", registry_dir=reg)
    engine.transition(job_id, "completed", registry_dir=reg, notify="completed")

    # The broker is unreachable, so the intent stays owed and flush retries it.
    notifier.flush(job_id, reg, attempts=1)
    assert notifier.pending(job_id, reg), "an unsent notification stays owed"
    notifier.flush(job_id, reg, attempts=1)

    payloads = [e["payload"] for e
                in mqtt_common.iter_logged_events(job_id, mqtt_common.get_logs_dir())
                if e.get("event") == "published"]
    assert len(payloads) == 2, "two send attempts for one notification"
    assert payloads[0]["seq"] != payloads[1]["seq"], (
        "the retry gets a fresh wire sequence: the monotonic guard cannot see it"
    )

    watcher = job_subscriber._Watcher({job_id}, {job_id: None},
                                      {job_id: payloads[0]["seq"] - 1})
    for payload in payloads:
        watcher.on_message(None, None, types.SimpleNamespace(
            topic="python/mqtt/jobs/%s/events" % job_id,
            payload=json.dumps(payload).encode()))
    assert watcher.events.qsize() == 1, "the same notification must be acted on once"