fix(ci): restore reviewed central Strix reliability contracts

.github/workflows/opencode-review.yml

@@ -573,25 +573,27 @@ jobs:
           if [ "$PR_DRAFT" = "true" ]; then
             echo "Event draft snapshot is stale; checking the verdict for the live ready PR."
           fi
-          reviews="$(gh api --paginate --slurp "repos/${TARGET_REPOSITORY}/pulls/${PR_NUMBER}/reviews?per_page=100")"
+          reviews_file="$(mktemp)"
           helper="$(mktemp)"
-          trap 'rm -f "$helper"' EXIT
+          trap 'rm -f "$reviews_file" "$helper"' EXIT
+          gh api --paginate --slurp "repos/${TARGET_REPOSITORY}/pulls/${PR_NUMBER}/reviews?per_page=100" >"$reviews_file"
           gh api "repos/ContextualWisdomLab/.github/contents/scripts/ci/opencode_review_receipt_gate.py?ref=${WORKFLOW_SHA}" \
             --jq .content | base64 --decode >"$helper"
-          verdict="$(python3 - "$helper" "$reviews" "$HEAD_SHA" <<'PY'
+          verdict="$(python3 - "$helper" "$reviews_file" "$HEAD_SHA" <<'PY'
           import importlib.machinery
           import importlib.util
           import json
           import sys
 
-          helper_path, reviews_json, head_sha = sys.argv[1:]
+          helper_path, reviews_path, head_sha = sys.argv[1:]
           loader = importlib.machinery.SourceFileLoader("trusted_opencode_receipt_gate", helper_path)
           spec = importlib.util.spec_from_loader(loader.name, loader)
           if spec is None or spec.loader is None:
               raise RuntimeError("trusted OpenCode receipt helper could not be loaded")
           gate = importlib.util.module_from_spec(spec)
           spec.loader.exec_module(gate)
-          pages = json.loads(reviews_json)
+          with open(reviews_path, encoding="utf-8") as stream:
+              pages = json.load(stream)
           if not isinstance(pages, list) or not all(
               isinstance(page, list) and all(isinstance(review, dict) for review in page)
               for page in pages

CHANGELOG.md

@@ -1,3 +1,16 @@
+### Strix report identity rejects colon-suffixed file lookalikes
+
+- The completed-report scope validator no longer treats every colon as proof
+  that a changed path token ended. A distinct Linux filename such as
+  `scripts/ci/strix_quick_gate.sh:backup` now fails closed, including beneath
+  the private Strix scope root, while numeric `path:line` and
+  `path:line:column` evidence remains valid. A sentence period also ends a
+  path token only before a real delimiter, so `strix_quick_gate.sh.:backup`
+  and `strix_quick_gate.sh.@backup` are rejected in both report forms.
+  Focused evidence is 69 tests across the three scope suites with 100%
+  statement and branch coverage for `strix_report_scope.py`; hosted
+  exact-head Checks and qualifying independent review remain required.
+
 ### Law CI cleanup rejects symlinked PostgreSQL descendants
 
 - The central law-agent cleanup now refuses symbolic links at every component

docs/product-technical-gap-baseline.md

@@ -71,10 +71,18 @@
 작성 기준일: **2026-08-26 10:35 KST**
 대상: **ContextualWisdomLab/.github** 중앙 거버넌스·자동화 레포지터리와 이를 소비하는 naruon 생태계
 현재 보호된 `main`: `826b92394c63deb6981c3a8d16a724d71f85a0d7`
-현재 열린 PR 수: **107** (아래 표에 이 스냅샷의 전체 목록 포함; live API 재수집)
+현재 열린 PR 수: **107** (아래 기준선 인벤토리 행 수; 이 스냅샷의 기계 계약)
+
+2026-10-07 delta: 그날 관측한 보호된 `main`은 `7554587c2e3106a388998bcad048a3d7121de25e`였고, live connector 조회는 반환 상한 100개를 모두 채웠으므로 열린 PR은 **최소 100개**였다. 이 관측값은 위 2026-08-26 스냅샷을 대체하지 않으며, 전체 수로 외삽하거나 아래 기준선 행 수를 재작성하지 않는다.
 
 이 문서는 제품·기술·운영 Gap을 현재 문서와 현재 GitHub 상태에 묶어 두는 기준선이다. 새 작업은 먼저 이 문서의 Gap ID를 PR 설명과 테스트 증거에 연결하고, PR의 정확한 exact HEAD·Checks·리뷰를 다시 수집한 뒤 구현한다. 표의 상태는 작성 시점의 관측값이므로, 병합 판단에는 재사용하지 않는다. 이 인벤토리는 스냅샷이며 merge authorization이 아니다.
 
+### 2026-10-07 Strix report-identity boundary delta
+
+| Gap ID | 상태 | exact-tree evidence | causal owner / next gate |
+|---|---|---|---|
+| CONTROL-STRIX-REPORT-PATH-IDENTITY-01 | **Source repaired on Draft `.github#2574`; protected integration pending** | RED `d8aeb759f780d58b13c8692fe52dd39cd4f60d80` proved that both `scripts/ci/strix_quick_gate.sh:backup` and its private-scope-root form were accepted as the changed file. GREEN `983e31ff828baeca9873042d8c3faf1b949f2df8` restricts colon suffixes to numeric `path:line[:column]` evidence. Focused evidence is 39 passed with 55/55 statements and 26/26 branches covered. Protected `main@7554587c2e3106a388998bcad048a3d7121de25e` was integrated by ordinary two-parent merge `d1cd94627279e1bcb9b5cd5fddbad9a0056d6e1a`. | Central `.github` owns the Strix report-to-changed-source identity boundary. Keep the PR Draft until its final exact head has hosted source-executing security/runtime Checks, no unresolved actionable review, and qualifying independent current-head approval. Do not treat zero-step runner failures as source evidence or use bypass. |
+
 ### 2026-09-30 central coverage owner stack delta
 
 | Gap ID | 상태 | exact-head evidence | causal owner / next gate |

scripts/ci/contextual_orchestrator_review_launcher.py

@@ -37,6 +37,7 @@
 
 from scripts.ci.contextual_orchestrator_review_policy import (
     FREE_POOL_CREDENTIAL_NAMES,
+    normalize_input_modalities,
     provider_account,
 )
 
@@ -261,6 +262,9 @@ def _report_rows(
                 "model": model_id,
                 "agent_id": str(getattr(model, "agent_id", None) or f"{provider}_{model_id}"),
                 "is_free": (provider, model_id) in free_route_identities,
+                "input_modalities": normalize_input_modalities(
+                    getattr(model, "input_modalities", None)
+                ),
                 "prompt_price_per_1k": getattr(model, "prompt_price_per_1k", None),
                 "completion_price_per_1k": getattr(model, "completion_price_per_1k", None),
                 "currency_code": getattr(model, "currency_code", None),

scripts/ci/strix_report_scope.py

@@ -10,10 +10,13 @@
 
 # The PR scan target is a private directory holding only the changed files, so
 # a path under it (as Strix reports it) is bound to this invocation's scope.
+# A sentence period ends a token only when a real delimiter (or end) follows it;
+# "name.:suffix" or "name.@suffix" is a different file, not the changed path.
+_TOKEN_DELIMITER = r"(?=$|[\s`\"'<>,;!?)])"
 PATH_TOKEN_END = (
     r"(?=$|[\s`\"'<>,;!?)]"
-    r"|:(?=[0-9]+(?::[0-9]+)?(?:$|[\s`\"'<>,;!?)]|\.(?![\w./-])))"
-    r"|\.(?![\w./-]))"
+    r"|:(?=[0-9]+(?::[0-9]+)?(?:$|[\s`\"'<>,;!?)]|\." + _TOKEN_DELIMITER + r"))"
+    r"|\." + _TOKEN_DELIMITER + r")"
 )
 SCOPE_PATH_RE = re.compile(
     r"(?<![\w./-])/workspace/strix-pr-scope\.[A-Za-z0-9]+/([A-Za-z0-9_./-]+)"

tests/test_codeql_ghas_configuration_identity.py

@@ -431,7 +431,8 @@ def test_request_json_maps_http_and_transport_failures(monkeypatch):
 
     class _HTTPError(identity.urllib.error.HTTPError):
         def read(self, size: int = -1) -> bytes:
-            return b"denied"[:size]
+            body = b"denied"
+            return body if size is None or size < 0 else body[:size]
 
     def raise_http(request, timeout=30):
         del request, timeout

tests/test_codeql_ghas_configuration_pagination.py

@@ -0,0 +1,140 @@
+"""Synthetic paginated HTTP inputs, not captured GitHub analyses.
+
+Pages advance through GitHub Link headers (main #2555 contract), not page counts.
+"""
+from urllib.parse import parse_qs, urlencode, urlsplit
+import pytest
+from scripts.ci import codeql_ghas_configuration_identity as identity
+
+BASE = 'a' * 40
+HEAD = 'b' * 40
+
+def _link_next(url, response_headers, has_next):
+    """Expose main's Link-header pagination contract for a synthetic page."""
+    if response_headers is None:
+        return
+    if not has_next:
+        response_headers['Link'] = ''
+        return
+    parts = urlsplit(url)
+    query = {key: values[0] for key, values in parse_qs(parts.query).items()}
+    query['page'] = str(int(query.get('page', '1')) + 1)
+    response_headers['Link'] = f'<{parts.scheme}://{parts.netloc}{parts.path}?{urlencode(query)}>; rel="next"'
+
+
+def row(sha, category='/language:actions'):
+    return {'commit_sha': sha, 'category': category, 'analysis_key': identity.DEFAULT_SETUP_ANALYSIS_KEY, 'tool': {'name': 'CodeQL'}}
+
+def test_base_identity_beyond_first_page_must_prevent_false_pairing_pass(monkeypatch):
+    calls = []
+    def request(url, *, token, timeout_seconds, response_headers=None):
+        params = parse_qs(urlsplit(url).query)
+        page = int(params.get('page', ['1'])[0])
+        calls.append(page)
+        assert params['ref'] == ['refs/heads/main']
+        _link_next(url, response_headers, page == 1)
+        if page == 1:
+            return [row('c' * 40, '/language:python') for _ in range(100)]
+        assert page == 2
+        return [row(BASE)]
+    monkeypatch.setattr(identity, '_request_json', request)
+    base = identity.list_codeql_analyses('ContextualWisdomLab/example', token='synthetic-fixture', ref='refs/heads/main')
+    ready, missing = identity.pairing_ready(base, [], base_sha=BASE, head_sha=HEAD, language='actions')
+    assert ready is False, 'truncated base inventory must not authorize pairing'
+    assert missing == [identity.default_setup_identity('actions')]
+    assert calls == [1, 2]
+
+
+def test_head_identity_on_later_page_satisfies_exact_pairing(monkeypatch):
+    calls = []
+    def request(url, *, token, timeout_seconds, response_headers=None):
+        page = int(parse_qs(urlsplit(url).query).get('page', ['1'])[0])
+        calls.append(page)
+        _link_next(url, response_headers, page == 1)
+        return [row('c' * 40) for _ in range(100)] if page == 1 else [row(HEAD)]
+    monkeypatch.setattr(identity, '_request_json', request)
+    head = identity.list_codeql_analyses('ContextualWisdomLab/example', token='synthetic-fixture', ref='refs/pull/1/head')
+    assert identity.pairing_ready([row(BASE)], head, base_sha=BASE, head_sha=HEAD, language='actions') == (True, [])
+    assert calls == [1, 2]
+
+
+@pytest.mark.parametrize('later', [{'malformed': True}, 'transport_error'])
+def test_later_page_error_cannot_return_partial_analyses(monkeypatch, later):
+    def request(url, *, token, timeout_seconds, response_headers=None):
+        page = int(parse_qs(urlsplit(url).query).get('page', ['1'])[0])
+        _link_next(url, response_headers, page == 1)
+        if page == 1:
+            return [row(BASE) for _ in range(100)]
+        if later == 'transport_error':
+            raise identity.ConfigurationIdentityError('synthetic later-page failure')
+        return later
+    monkeypatch.setattr(identity, '_request_json', request)
+    with pytest.raises(identity.ConfigurationIdentityError):
+        identity.list_codeql_analyses('ContextualWisdomLab/example', token='synthetic-fixture')
+
+
+@pytest.mark.parametrize('per_page', [0, -1, 101, True, 1.5, '100'])
+def test_invalid_page_size_rejected_before_request(monkeypatch, per_page):
+    calls = []
+    def request(*args, **kwargs):
+        calls.append(True)
+        raise RuntimeError('invalid page size reached transport')
+    monkeypatch.setattr(identity, '_request_json', request)
+    with pytest.raises(identity.ConfigurationIdentityError):
+        identity.list_codeql_analyses('ContextualWisdomLab/example', token='synthetic-fixture', per_page=per_page)
+    assert calls == []
+
+
+def test_full_page_filtered_rows_still_requires_next_page(monkeypatch):
+    calls = []
+    def request(url, *, token, timeout_seconds, response_headers=None):
+        page = int(parse_qs(urlsplit(url).query).get('page', ['1'])[0])
+        calls.append(page)
+        _link_next(url, response_headers, page == 1)
+        return [None for _ in range(100)] if page == 1 else [row(BASE)]
+    monkeypatch.setattr(identity, '_request_json', request)
+    assert identity.list_codeql_analyses('ContextualWisdomLab/example', token='synthetic-fixture') == [row(BASE)]
+    assert calls == [1, 2]
+
+@pytest.mark.parametrize("boundary", ["codeql-body", "launcher-close", "launcher-long-header"])
+def test_http_exception_custody_and_bounded_classification(monkeypatch, boundary):
+    """Real modules must bound HTTP diagnostics and settle error streams offline."""
+    import io
+    import urllib.error
+    from scripts.ci import contextual_orchestrator_review_launcher as launcher
+    from types import SimpleNamespace
+
+    class Stream(io.BytesIO):
+        def __init__(self):
+            super().__init__(b"synthetic diagnostic " * 100)
+            self.read_sizes = []
+
+        def read(self, size=-1):
+            self.read_sizes.append(size)
+            return super().read(size)
+
+    stream = Stream()
+    error = urllib.error.HTTPError("https://api.github.com/fixture", 429, "synthetic", {"Retry-After": "12"}, stream)
+    try:
+        if boundary == "codeql-body":
+            def refuse(*args, **kwargs):
+                raise error
+            monkeypatch.setattr(identity._GITHUB_API_OPENER, "open", refuse)
+            with pytest.raises(identity.ConfigurationIdentityError):
+                identity._request_json("https://api.github.com/fixture", token="synthetic-only", timeout_seconds=1)
+            assert stream.read_sizes == [400], "HTTP diagnostic read must be bounded"
+            assert stream.closed, "HTTP error must close before propagation"
+        elif boundary == "launcher-close":
+            row = {"finish_reason": "stale", "reasoning_without_content": True}
+            launcher._record_provider_exception(row, error)
+            assert row == {"status": "rejected", "error_type": "HTTPError", "http_status": 429, "retry_after_s": 12}
+            assert stream.closed, "provider HTTP error must close after classification"
+        else:
+            try:
+                value = launcher._safe_retry_after_seconds(SimpleNamespace(headers={"Retry-After": "9" * 5000}))
+            except ValueError:
+                pytest.fail("long decimal header escaped safe classification")
+            assert value is None
+            assert launcher._safe_retry_after_seconds(error) == 12
+    finally:
+        error.close()

tests/test_contextual_orchestrator_review_policy.py

@@ -796,3 +796,30 @@ def test_build_catalog_interleaving_skips_exhausted_accounts() -> None:
         "nvidia_nim",
         "nvidia_nim",
     ]
+
+
+@pytest.mark.parametrize("value, expected", [
+    (None, []),
+    ([], []),
+    ((" TEXT ", "image", "text", "audio_2", "video-1"),
+     ["text", "image", "audio_2", "video-1"]),
+])
+def test_valid_modality_evidence_retains_order_and_deduplicates(value, expected):
+    """Missing evidence stays empty; valid explicit evidence is preserved."""
+    assert policy.normalize_input_modalities(value) == expected
+
+
+@pytest.mark.parametrize("value, message", [
+    ("text", "must be a list or tuple"),
+    ({"text": True}, "must be a list or tuple"),
+    ([False], "non-string token"),
+    (["text", 1], "non-string token"),
+    ([""], "invalid token"),
+    (["image extra"], "invalid token"),
+    (["../image"], "invalid token"),
+    (["追加"], "invalid token"),
+])
+def test_malformed_evidence_fails_closed(value, message):
+    """Malformed evidence cannot lose filtering tags and become a valid model."""
+    with pytest.raises(policy.PolicyError, match=message):
+        policy.normalize_input_modalities(value)

tests/test_migrated_receipt_rejection.py

@@ -0,0 +1,17 @@
+"""Explicit invalid receipts must stay invalid after test-harness migration."""
+import pytest
+from tests.test_opencode_required_verdict_regression import HEAD, runtime_verdict
+
+
+@pytest.mark.parametrize('overrides', [
+    {'body': '## OpenCode Review Overview\n- Gate result: COMMENT'},
+    {'body': '@opencode-agent please review'},
+    {'body': '## Pull request overview\n- Head SHA: `' + 'b' * 40 + '`'},
+    {'id': None},
+])
+def test_invalid_receipt_is_not_improved_by_fixture_constructor(overrides):
+    row = {'id': 1, 'state': 'APPROVED', 'commit_id': HEAD,
+           'user': {'login': 'opencode-agent[bot]'},
+           'body': '## Pull request overview\n- Head SHA: `' + HEAD + '`'}
+    row.update(overrides)
+    assert runtime_verdict([row]) == ''

tests/test_opencode_required_rerun_capacity.py

@@ -105,6 +105,8 @@ def test_missing_verdict_fails_after_one_review_read(tmp_path: Path) -> None:
 
     assert result.returncode == 1
     assert "will rerun this failed job" in result.stdout
+    assert "No current-head formal OpenCode verdict:" in result.stderr
+    assert "Current-head OpenCode verdict:" not in result.stdout
     assert calls.read_text(encoding="utf-8").splitlines() == [
         "api repos/owner/repo/pulls/7",
         "api --paginate --slurp repos/owner/repo/pulls/7/reviews?per_page=100",

tests/test_opencode_required_verdict_regression.py

@@ -422,6 +422,7 @@ def _run_fail_closed_step(
             "HEAD_SHA": head_sha,
             "PR_ACTION": pr_action,
             "PR_DRAFT": pr_draft,
+            "WORKFLOW_SHA": "6a37e4cdfbd8bf6f3a60645a0ad7c13e1b9c1ae3",
             "LIVE_PR_JSON": json.dumps(
                 {
                     "draft": pr_draft.lower() == "true",