code-coordinator 0.5.522__tar.gz → 0.5.524__tar.gz

{code_coordinator-0.5.522 → code_coordinator-0.5.524}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.522
+Version: 0.5.524
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.522 → code_coordinator-0.5.524}/code_coordinator.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.522
+Version: 0.5.524
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.522 → code_coordinator-0.5.524}/code_coordinator.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "0.5.522",
+  "tag": "0.5.524",
   "distance": 0,
-  "node": "g3d51cf0a782a606306c6175a5b2271d2d93ea2d4",
+  "node": "g0e924a9e07fea9e3ea943a45cd49b370575ca40c",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-09-27"

{code_coordinator-0.5.522 → code_coordinator-0.5.524}/coord/config.py

@@ -307,12 +307,34 @@ class SmokeTestsConfig:
     `default_command` is the shell command the smoke agent runs (e.g.
     `make smoke` or `pytest tests/smoke`). Per-repo overrides flow through
     `Repo.test_command` already; this is the fallback when none is set.
+
+    `native_execution_capabilities` (#3455) lists which bare capabilities
+    (e.g. `"windows"`) merely mean "some machine can build/cross-compile
+    for this" and do NOT by themselves prove a Test leg runs on that OS
+    natively. `dell64` declared `windows` while its coord agent ran inside
+    WSL (Linux); a #[cfg(target_os = "windows")] test compiled out and the
+    Test stage reported `pass` for a failure GitHub's `windows-latest`
+    runner caught every time (quadraui#1077). A machine must ALSO declare
+    `coord.config.native_execution_capability(cap)` (e.g. `windows-native`)
+    to certify it runs *that* capability's suite natively — see
+    `coord.smoke.unverified_native_capabilities`, consulted at THREE points
+    so the gap can't be closed by prompt-cooperation alone: routing
+    (`coord.smoke.rank_smoke_machines` prefers a native-verified candidate
+    over a bare one when both exist), dispatch (the briefing tells the
+    worker to caveat its own marker), and verdict-recording
+    (`coord.notify._record_smoke_verdict`'s PASS branches independently
+    recompute this and record `test_confirmation="native_unverified"`
+    regardless of what the worker's marker text says). Defaults to
+    `["windows"]`, the one capability the incident this exists for actually
+    involved; add more here (e.g. a future `"macos"` cross-build path)
+    without any code change.
     """
 
     auto_queue: bool = False
     default_command: str | None = None
     timeout_seconds: int = 600
     capability_rules: list[SmokeRule] = field(default_factory=list)
+    native_execution_capabilities: list[str] = field(default_factory=lambda: ["windows"])
 
 
 @dataclass
@@ -1904,6 +1926,25 @@ def provider_capability(provider_type: str) -> str:
     return f"provider:{provider_type}"
 
 
+def native_execution_capability(capability: str) -> str:
+    """The ``capabilities:`` string a machine must ALSO declare, alongside
+    *capability* itself, to certify it runs *capability*'s tests on a
+    genuinely native host rather than merely building/cross-compiling for
+    it (#3455).
+
+    ``native_execution_capability("windows") == "windows-native"``. Same
+    config-only-declaration shape as :func:`provider_capability` — a
+    machine that lists ``windows`` in ``capabilities:`` can build for
+    Windows (a WSL/Linux agent satisfies that today), but only one that
+    ALSO lists ``windows-native`` has told the coordinator its Test-stage
+    runs actually execute there. Single source of truth for the naming
+    convention: `coord.smoke.unverified_native_capabilities` is the only
+    caller, so the two can never disagree about the string (#2096, "one
+    question, one answer").
+    """
+    return f"{capability}-native"
+
+
 def model_plausible_for_provider_type(model: str, provider_type: str) -> bool:
     """Namespace-shape sanity check: could *model* plausibly belong to
     *provider_type* (#1798)?
@@ -3498,6 +3539,14 @@ def _parse_smoke_tests(raw: Any) -> SmokeTestsConfig:
             raise ConfigError("smoke_tests.timeout_seconds must be a positive integer")
         cfg.timeout_seconds = value
 
+    if "native_execution_capabilities" in raw:
+        value = raw["native_execution_capabilities"]
+        if not isinstance(value, list) or not all(isinstance(c, str) for c in value):
+            raise ConfigError(
+                "smoke_tests.native_execution_capabilities must be a list of strings"
+            )
+        cfg.native_execution_capabilities = value
+
     rules_raw = raw.get("capability_rules", []) or []
     if not isinstance(rules_raw, list):
         raise ConfigError("smoke_tests.capability_rules must be a list")

{code_coordinator-0.5.522 → code_coordinator-0.5.524}/coord/confirm_test.py

@@ -314,8 +314,21 @@ INCONCLUSIVE_KINDS = frozenset({
 # * ``TEST_CONFIRMATION_BASELINE_RED`` — `ConfirmationResult.baseline_red`
 #   (#2170): the suite failed identically on the merge-base, so nothing was
 #   learned about the BRANCH specifically, but a real run did happen.
+# * ``TEST_CONFIRMATION_NATIVE_UNVERIFIED`` (#3455) — the claimed pass ran on
+#   a machine that declares only the bare/build side of a capability this
+#   leg required (`coord.smoke.SmokeTestsConfig.native_execution_
+#   capabilities`, e.g. `"windows"`), not the paired `coord.config.
+#   native_execution_capability` proof it runs that suite on genuinely native
+#   hardware — a WSL-hosted `windows` machine being the canonical case
+#   (`dell64` / quadraui#1077). Distinct from `UNCONFIRMED`: a confirmation
+#   run may well have executed and even agreed with the claim, but "agreed"
+#   here still means "on a non-native host", so platform-conditional code the
+#   real target OS would exercise never ran anywhere in this leg. Recorded
+#   by `coord.notify._record_smoke_verdict`'s PASS branches regardless of
+#   what the worker's own `SMOKE:` marker text says — see `coord.smoke.
+#   native_unverified_for_verdict`.
 #
-# ``None`` (the column default) is a fifth, implicit state: no confirmation
+# ``None`` (the column default) is a sixth, implicit state: no confirmation
 # question was ever asked about this write at all — a headless smoke
 # FAILURE, a human `coord test --passed` write that has not yet been reaped
 # by a notify pass, a mute-leg park, ``TEST_STATE_BLOCKED``, ... None of
@@ -326,6 +339,7 @@ TEST_CONFIRMATION_CONFIRMED = "confirmed"
 TEST_CONFIRMATION_UNCONFIRMED = "unconfirmed"
 TEST_CONFIRMATION_REFUTED = "refuted"
 TEST_CONFIRMATION_BASELINE_RED = "baseline_red"
+TEST_CONFIRMATION_NATIVE_UNVERIFIED = "native_unverified"
 
 #: Every value `test_confirmation` may hold — for validation / display sites
 #: that want to assert exhaustiveness rather than silently falling through.
@@ -334,6 +348,7 @@ TEST_CONFIRMATION_VALUES = frozenset({
     TEST_CONFIRMATION_UNCONFIRMED,
     TEST_CONFIRMATION_REFUTED,
     TEST_CONFIRMATION_BASELINE_RED,
+    TEST_CONFIRMATION_NATIVE_UNVERIFIED,
 })
 
 #: #3378: the canonical prefix a ``test_reason`` carries when a ``skipped``
@@ -1464,6 +1479,7 @@ __all__ = [
     "STALE_WORKTREE_MAX_AGE_HOURS",
     "TEST_CONFIRMATION_BASELINE_RED",
     "TEST_CONFIRMATION_CONFIRMED",
+    "TEST_CONFIRMATION_NATIVE_UNVERIFIED",
     "TEST_CONFIRMATION_REFUTED",
     "TEST_CONFIRMATION_UNCONFIRMED",
     "TEST_CONFIRMATION_VALUES",