claude-dev-env 8.82.1 → 8.83.1

package/scripts/dev_env_scripts_constants/test_account_broker_constants.py

@@ -191,6 +191,21 @@ def test_should_treat_accounts_that_differ_only_in_profile_as_one_account() -> N
     assert Account(Product.CLAUDE, "first", home, profile="first") == Account(Product.CLAUDE, "first", home)
 
 
+def test_should_wait_until_a_rate_limited_meter_may_be_probed_again() -> None:
+    now = datetime(2026, 10, 11, 19, 31, tzinfo=timezone.utc)
+    probe_again_at = now + timedelta(minutes=5)
+    account = Account(Product.CLAUDE, "second", Path("second"))
+    reading = broker_constants.Reading(account, None, "usage probe failed: HTTP Error 429", None, probe_again_at)
+
+    decision = broker_constants._wait_decision([reading], now, frozenset(), {})
+
+    assert decision.resets_at == probe_again_at
+    assert decision.reason == (
+        "no account meter could be read (second (usage probe failed: HTTP Error 429)); "
+        "next check at 2026-10-11T19:36:00+00:00"
+    )
+
+
 def _codex_reading(meters: broker_constants.Meters) -> broker_constants.Reading:
     return broker_constants.Reading(Account(Product.CODEX, "codex-1", Path("/profiles/codex-1")), meters)
 

package/scripts/policy_lint/adapter_configuration.py

@@ -135,6 +135,7 @@ def _names_exempt_registration_path(registered_string: str) -> bool:
         hooks/blocking/verify_before_acting.py           -> exempt
         hooks/blocking/session_title_format_gate.py      -> exempt
         hooks/blocking/artifact_dark_mode_gate.py        -> exempt
+        hooks/blocking/build_eval_before_eval_write_gate.py -> exempt
         hooks/blocking/merge_over_review_offer_gate.py   -> exempt
         hooks/blocking/final_reply_gate.py               -> exempt
         hooks/blocking/some_new_blocker.py               -> flagged
@@ -168,6 +169,10 @@ def _names_exempt_registration_path(registered_string: str) -> bool:
     The artifact page gate denies a page publish whose text is unreadable in
     dark mode. It decides how a page renders and no code or safety policy.
 
+    The eval write gate denies a write to an eval file until the session
+    invokes ``/claude-api build-eval``. It holds eval writes until a skill
+    loads, and it judges no code or safety content.
+
     The merge offer gate denies a decision card whose text offers a merge and
     names open review findings. It reads only the wording of a chat card and
     decides no code or safety policy; the merge review gate holds the merge.

package/scripts/policy_lint/config/constants.py

@@ -74,6 +74,7 @@ ALL_ACTION_BOUNDARY_EXEMPT_REGISTRATION_PATHS = (
     "blocking/image_read_size_gate.py",
     "blocking/artifact_dark_mode_gate.py",
     "blocking/haiku_spawn_slim_gate.py",
+    "blocking/build_eval_before_eval_write_gate.py",
     "blocking/merge_over_review_offer_gate.py",
     "blocking/final_reply_gate.py",
 )

package/scripts/test_account_broker.py

@@ -470,12 +470,30 @@ def test_should_count_failed_meter_read_as_spent() -> None:
 
 
 def test_should_serialize_readings_with_named_fields() -> None:
-    reading = Reading(_account("one"), _meters(80, 70))
+    reading = Reading(_account("one"), _meters(80, 70), measured_at=NOW)
 
     payload = account_broker.readings_payload((reading,))
 
     assert payload[0]["name"] == "one"
     assert payload[0]["meters"]["weekly_percent_left"] == 70
+    assert payload[0]["measured_at"] == NOW.isoformat()
+
+
+def test_should_wait_until_the_rate_limit_probe_after_stale_meters_expire() -> None:
+    reading = Reading(
+        _account("one", Product.CLAUDE), None,
+        "usage probe failed: HTTP Error 429", NOW - timedelta(minutes=25),
+        NOW + timedelta(minutes=5),
+    )
+
+    decision = choose_from_readings(Product.CLAUDE, (reading,), now=NOW)
+
+    assert decision.action == "wait"
+    assert decision.resets_at == NOW + timedelta(minutes=5)
+    assert decision.reason == (
+        "no account meter could be read (one (usage probe failed: HTTP Error 429)); "
+        "next check at 2026-10-03T00:05:00+00:00"
+    )
 
 
 def test_should_wait_when_all_meter_reads_fail() -> None:

package/scripts/test_account_broker_support.py

@@ -7,7 +7,7 @@ import shutil
 import subprocess
 import sys
 import time
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 from collections.abc import Callable
 from pathlib import Path
 from types import SimpleNamespace
@@ -341,10 +341,11 @@ def test_should_cap_the_meter_read_delay(monkeypatch: pytest.MonkeyPatch, tmp_pa
     all_failures = [support.WeeklyUtilizationProbeError("down") for _ in range(6)]
     _claude_meter_reads(monkeypatch, all_failures)
 
-    meters, unread_reason = support.read_meters_with_backoff(support.read_claude_meters, Account(Product.CLAUDE, "one", tmp_path))
+    meters, unread_reason, is_rate_limited = support.read_meters_with_backoff(support.read_claude_meters, Account(Product.CLAUDE, "one", tmp_path))
 
     assert meters is None
     assert unread_reason == "down"
+    assert not is_rate_limited
     assert all_delays == [1.0, 2.0, 4.0, 8.0, 8.0]
 
 
@@ -356,10 +357,11 @@ def test_should_read_a_signed_out_claude_account_once(monkeypatch: pytest.Monkey
         all_reads.append(account)
         raise support.ClaudeSignedOutError("no usable bearer token")
 
-    meters, unread_reason = support.read_meters_with_backoff(signed_out_reader, Account(Product.CLAUDE, "main", tmp_path))
+    meters, unread_reason, is_rate_limited = support.read_meters_with_backoff(signed_out_reader, Account(Product.CLAUDE, "main", tmp_path))
 
     assert meters is None
     assert unread_reason == "no usable bearer token"
+    assert not is_rate_limited
     assert len(all_reads) == 1
     assert all_delays == []
 
@@ -385,6 +387,162 @@ def test_should_cache_account_reading_under_injected_state_path(
     assert support.broker_state_path().is_file()
 
 
+def _read_twice(monkeypatch: pytest.MonkeyPatch, tmp_path: Path, all_results: list[object], gap: timedelta) -> object:
+    monkeypatch.setattr(support, "broker_state_path", lambda: tmp_path / "broker" / "state.json")
+    _recorded_sleeps(monkeypatch)
+    account = Account(Product.CLAUDE, "second", tmp_path)
+    pending = list(all_results)
+
+    def reader(_selected: Account) -> Meters:
+        next_outcome = pending.pop(0)
+        if isinstance(next_outcome, Exception):
+            raise next_outcome
+        return next_outcome
+
+    adapter = ProductAdapter(lambda: (account,), reader, "CLAUDE_CONFIG_DIR", ())
+    state = support._load_state(support.broker_state_path())
+    first_read_at = datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+    support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first_read_at)
+    return support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first_read_at + gap)[0]
+
+
+def test_should_keep_the_last_good_meters_when_the_usage_endpoint_rate_limits(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    rate_limited = support.WeeklyUtilizationProbeError("usage probe failed: HTTP Error 429: Too Many Requests")
+    reading = _read_twice(monkeypatch, tmp_path, [Meters(74.0, None, 81.0, None), rate_limited], timedelta(minutes=5))
+
+    assert reading.meters == Meters(74.0, None, 81.0, None)
+    assert reading.unread_reason == str(rate_limited)
+    assert reading.measured_at == datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+
+
+def test_should_drop_last_good_meters_older_than_the_fallback_window(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    rate_limited = support.WeeklyUtilizationProbeError("usage probe failed: HTTP Error 429: Too Many Requests")
+    reading = _read_twice(monkeypatch, tmp_path, [Meters(74.0, None, 81.0, None), rate_limited], timedelta(hours=2))
+
+    assert reading.meters is None
+    assert reading.unread_reason == "usage probe failed: HTTP Error 429: Too Many Requests"
+
+
+def test_should_keep_last_good_meters_at_the_fallback_boundary(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    rate_limited = support.WeeklyUtilizationProbeError("usage probe failed: HTTP Error 429")
+    reading = _read_twice(
+        monkeypatch, tmp_path,
+        [Meters(74.0, None, 81.0, None), rate_limited], timedelta(minutes=30),
+    )
+
+    assert reading.meters == Meters(74.0, None, 81.0, None)
+
+
+def test_should_drop_last_good_meters_after_a_window_reset(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    first_read = datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+    rate_limited = support.WeeklyUtilizationProbeError("usage probe failed: HTTP Error 429")
+    reading = _read_twice(
+        monkeypatch, tmp_path,
+        [Meters(3.0, first_read + timedelta(minutes=10), 81.0, None), rate_limited],
+        timedelta(minutes=25),
+    )
+
+    assert reading.meters is None
+    assert reading.unread_reason == str(rate_limited)
+
+
+def test_should_drop_held_meters_when_a_window_resets(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    monkeypatch.setattr(support, "broker_state_path", lambda: tmp_path / "broker" / "state.json")
+    account = Account(Product.CLAUDE, "second", tmp_path)
+    start = datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+    state = support._load_state(support.broker_state_path())
+    state["meters"][support._state_key(account)] = {
+        "read_at": start.timestamp(),
+        "measured_at": (start - timedelta(minutes=5)).timestamp(),
+        "meters": support._meter_payload(Meters(3.0, start + timedelta(minutes=1), 81.0, None)),
+        "unread_reason": "usage probe failed: HTTP Error 429",
+        "retry_at": (start + timedelta(minutes=5)).timestamp(),
+    }
+    adapter = ProductAdapter(lambda: (account,), lambda _: pytest.fail("probe during cooldown"), "CLAUDE_CONFIG_DIR", ())
+
+    reading = support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=start + timedelta(minutes=2))[0]
+
+    assert reading.meters is None
+    assert reading.retry_at == start + timedelta(minutes=5)
+
+
+def test_should_reject_a_last_good_entry_with_an_unrelated_failure() -> None:
+    now = datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+    entry = {
+        "measured_at": now.timestamp(),
+        "meters": support._meter_payload(Meters(74.0, None, 81.0, None)),
+        "unread_reason": "credentials file missing",
+    }
+
+    assert support._last_good_meters(entry, now) is None
+
+
+def test_should_not_use_last_good_meters_after_a_file_failure(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    reading = _read_twice(
+        monkeypatch, tmp_path,
+        [Meters(74.0, None, 81.0, None), *[OSError("credentials file missing") for _ in range(3)]],
+        timedelta(minutes=5),
+    )
+
+    assert reading.meters is None
+    assert reading.unread_reason == "credentials file missing"
+
+
+def test_should_keep_first_measurement_time_across_rate_limit_fallbacks(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    monkeypatch.setattr(support, "broker_state_path", lambda: tmp_path / "broker" / "state.json")
+    _recorded_sleeps(monkeypatch)
+    account = Account(Product.CLAUDE, "second", tmp_path)
+    rate_limited = support.WeeklyUtilizationProbeError("usage probe failed: HTTP Error 429")
+    pending = [Meters(74.0, None, 81.0, None), rate_limited, rate_limited]
+    calls = []
+
+    def reader(_selected: Account) -> Meters:
+        calls.append(1)
+        outcome = pending.pop(0)
+        if isinstance(outcome, Exception):
+            raise outcome
+        return outcome
+
+    adapter = ProductAdapter(lambda: (account,), reader, "CLAUDE_CONFIG_DIR", ())
+    state = support._load_state(support.broker_state_path())
+    first = datetime(2026, 10, 11, 19, 20, tzinfo=timezone.utc)
+    support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first)
+    second = support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first + timedelta(minutes=5))[0]
+    held = support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first + timedelta(minutes=6))[0]
+    third = support.read_accounts(Product.CLAUDE, adapter, all_state=state, now=first + timedelta(minutes=10))[0]
+
+    assert second.measured_at == first
+    assert held.measured_at == first
+    assert third.measured_at == first
+    assert len(calls) == 3
+
+
+def test_should_keep_newer_measurement_when_stale_fallback_saves_last(tmp_path: Path) -> None:
+    state_path = tmp_path / "broker" / "state.json"
+    account = Account(Product.CLAUDE, "second", tmp_path)
+    key = support._state_key(account)
+    old = {"meters": {key: {"read_at": 40.0, "measured_at": 10.0, "meters": support._meter_payload(Meters(74.0, None, 81.0, None)), "unread_reason": "HTTP Error 429"}}, "spent": {}, "affinity": {}}
+    newer = {"meters": {key: {"read_at": 30.0, "measured_at": 30.0, "meters": support._meter_payload(Meters(5.0, None, 81.0, None)), "unread_reason": None}}, "spent": {}, "affinity": {}}
+    support._save_state(state_path, newer)
+    support._save_state(state_path, old)
+
+    assert support._load_state(state_path)["meters"][key]["meters"]["session_percent_left"] == 5.0
+
+
 def test_should_extract_session_id_and_restore_runner() -> None:
     original = support.subprocess_runner
 

package/scripts/tests/test_adapter_configuration.py

@@ -130,6 +130,14 @@ def test_new_artifact_page_gate_registration_is_clean(tmp_path: Path) -> None:
     assert adapters.hook_configuration_diagnostics(current_document, tmp_path) == ()
 
 
+def test_new_eval_write_gate_registration_is_clean(tmp_path: Path) -> None:
+    current_document = _hook_document(
+        ["hooks/blocking/build_eval_before_eval_write_gate.py"],
+        '{"hooks": {}}',
+    )
+    assert adapters.hook_configuration_diagnostics(current_document, tmp_path) == ()
+
+
 def test_new_blocker_in_staged_change_is_rejected(tmp_path: Path) -> None:
     current_document = _hook_document(
         ["hooks/blocking/new.py"],

package/scripts/tests/test_rule_load_scopes.py

@@ -8,6 +8,7 @@ SKILLS_DIRECTORY = PACKAGE_ROOT / ".agents" / "skills"
 
 ALWAYS_ON_RULE_NAMES = frozenset(
     {
+        "agent-board.md",
         "correction-lens.md",
         "index.md",
         "question-presentation.md",