fix(noema): bounded complete serializer prerequisite

.github/workflows/agent-review-runtime-quality-ci.yml

@@ -67,13 +67,15 @@ on:
       - "scripts/ci/zdr_policy.py"
       - "scripts/ci/contextual_orchestrator_review_policy.py"
       - "scripts/ci/contextual_orchestrator_review_launcher.py"
+      - "scripts/ci/noema_review_gate.py"
       - "scripts/ci/personal_review_route.py"
       - "scripts/ci/contextual_orchestrator_review_sidecar.sh"
       - "tests/test_zdr_policy.py"
       - "tests/test_contextual_orchestrator_review_policy.py"
       - "tests/test_contextual_orchestrator_review_launcher.py"
       - "tests/test_review_auto_route.py"
       - "tests/test_personal_review_route.py"
+      - "tests/test_noema_complete_auto_input.py"
       - "tests/test_tokyo_review_consumer.py"
       - "tests/test_review_auto_cost_admission.py"
       - "tests/test_review_ai_app_independence.py"
@@ -221,6 +223,10 @@ jobs:
                 noema_suite=true
                 review_repair_suite=true
                 ;;
+              scripts/ci/noema_review_gate.py)
+                noema_suite=true
+                review_repair_suite=true
+                ;;
               .github/workflows/noema-review.yml|\
               .github/actions/noema-review/two_phase.py|\
               tests/test_noema_reviewer_token_lifetime.py|\
@@ -304,6 +310,7 @@ jobs:
               tests/test_contextual_orchestrator_review_launcher.py|\
               tests/test_review_auto_route.py|\
               tests/test_personal_review_route.py|\
+              tests/test_noema_complete_auto_input.py|\
               tests/test_tokyo_review_consumer.py|\
               tests/test_review_auto_cost_admission.py|\
               tests/test_review_ai_app_independence.py|\
@@ -504,6 +511,7 @@ jobs:
             tests/test_contextual_orchestrator_review_launcher.py \
             tests/test_review_auto_route.py \
             tests/test_personal_review_route.py \
+            tests/test_noema_complete_auto_input.py \
             tests/test_tokyo_review_consumer.py \
             tests/test_review_auto_cost_admission.py \
             tests/test_review_ai_app_independence.py \
@@ -550,6 +558,7 @@ jobs:
             tests/test_contextual_orchestrator_review_launcher.py \
             tests/test_review_auto_route.py \
             tests/test_personal_review_route.py \
+            tests/test_noema_complete_auto_input.py \
             tests/test_tokyo_review_consumer.py \
             tests/test_review_auto_cost_admission.py \
             tests/test_review_ai_app_independence.py \

scripts/ci/noema_review_gate.py

@@ -66,6 +66,20 @@
 # noema_review_handoff.py.
 NOEMA_REVIEW_BODY_HEAD_RE = re.compile(r"^- Head SHA:\s*`([0-9a-fA-F]{40})`$", re.MULTILINE)
 MAX_DIFF_CHARS = 60000
+# Personal auto advertises 500,000 input tokens. Bound the complete JSON
+# request conservatively by UTF-8 bytes (including schema/envelope), reserving
+# 8,192 for chat framing. Do not apply this route's capacity to other models.
+MAX_AUTO_REVIEW_REQUEST_BYTES = 500000 - 8192
+
+
+def complete_auto_input() -> bool:
+    """Select only the existing personal-auto endpoint, not arbitrary aliases."""
+    return (
+        os.environ.get("NOEMA_LLM_MODEL", "").strip() == "auto"
+        and os.environ.get("NOEMA_LLM_API_URL", "").strip()
+        == "https://litellm.poinnetworks.net/v1/chat/completions"
+    )
+
 MAX_CONTEXT_FILES = 12
 MAX_FILE_CONTEXT_CHARS = 4000
 MAX_REVIEW_CONTEXT_CHARS = 24000
@@ -542,6 +556,10 @@ def require_distinct_pr_author(pr: dict[str, Any], actor: str) -> str:
 def fetch_diff(repo: str, number: int) -> tuple[str, bool]:
     """Fetch the PR diff and truncate it to the bounded LLM prompt size."""
     diff = run(["gh", "api", f"repos/{repo}/pulls/{number}", "-H", "Accept: application/vnd.github.v3.diff"])
+    if complete_auto_input():
+        if len(diff.encode("utf-8")) > MAX_AUTO_REVIEW_REQUEST_BYTES:
+            raise RuntimeError("Noema complete review input exceeds personal auto budget")
+        return diff, False
     truncated = len(diff) > MAX_DIFF_CHARS
     if truncated:
         marker = "[overlong changed line content omitted]"
@@ -1625,6 +1643,8 @@ def call_llm(
             "Noema LLM review unavailable: NOEMA_LLM_API_URL or NOEMA_LLM_API_KEY is not configured."
         )
     reject_private_llm_url(api_url)
+    if complete_auto_input() and truncated:
+        raise RuntimeError("Noema complete review input requires an untruncated diff")
 
     allowed_locations = [
         {"path": path, "line": line, "side": side}
@@ -1633,7 +1653,10 @@ def call_llm(
     location_example = allowed_locations[0] if allowed_locations else {
         "path": "path", "line": 0, "side": "RIGHT"
     }
-    allowed_locations_json = _bounded_allowed_locations_json(allowed_locations)
+    allowed_locations_json = (
+        json.dumps({"locations": allowed_locations, "truncated": False}, separators=(",", ":"))
+        if complete_auto_input() else _bounded_allowed_locations_json(allowed_locations)
+    )
     prompt = {
         "role": "user",
         "content": "\n".join(
@@ -1668,9 +1691,12 @@ def call_llm(
             prompt,
         ],
     }
+    payload_bytes = json.dumps(payload).encode("utf-8")
+    if complete_auto_input() and len(payload_bytes) > MAX_AUTO_REVIEW_REQUEST_BYTES:
+        raise RuntimeError("Noema complete review input exceeds personal auto budget")
     request = urllib.request.Request(
         api_url,
-        data=json.dumps(payload).encode("utf-8"),
+        data=payload_bytes,
         headers={
             "authorization": f"Bearer {api_key}",
             "content-type": "application/json",

tests/test_agent_review_runtime_quality_consolidation.py

@@ -148,6 +148,8 @@ def test_ghas_runtime_changes_start_the_quality_runner(changed_path: str) -> Non
 @pytest.mark.parametrize(
     ("changed_path", "expected_suites"),
     (
+        ("tests/test_noema_complete_auto_input.py", {"review_repair"}),
+        ("scripts/ci/noema_review_gate.py", {"noema", "review_repair"}),
         ("scripts/ci/codeql_ghas_configuration_identity.py", {"ghas"}),
         ("tests/test_codeql_ghas_configuration_identity.py", {"ghas"}),
         ("tests/test_codeql_ghas_analyses_pagination.py", {"ghas"}),
@@ -385,3 +387,17 @@ def test_exact_artifact_suite_preserves_version_and_quality_contracts() -> None:
     assert "outputs.exact_artifact == 'true'" in workflow
     assert "--include=scripts/ci/verify_exact_artifact_sbom_handoff.py" in workflow
     assert "interrogate --fail-under=100" in workflow
+
+
+def test_complete_serializer_enrollment():
+    workflow = _workflow_text()
+    path = "tests/test_noema_complete_auto_input.py"
+    trigger = workflow.split("on:\n", 1)[1].split("\nconcurrency:\n", 1)[0]
+    assert f'      - "{path}"' in trigger
+    assert workflow.count(path) == 4
+    step = next(part for part in workflow.split('      - name: ')
+                if 'python -m pytest -q' in part and path in part)
+    pytest_args = step.split('python -m pytest -q', 1)[1].split('python -m interrogate', 1)[0]
+    compile_args = step.split('python -m compileall -q', 1)[1]
+    assert pytest_args.count(path) == 1
+    assert compile_args.count(path) == 1

tests/test_noema_complete_auto_input.py

@@ -0,0 +1,70 @@
+"""Complete source admission through the real Noema serializer (no network)."""
+import json
+from unittest.mock import patch
+import pytest
+from scripts.ci import noema_review_gate as gate
+
+
+def test_real_serializer_preserves_all_locations(monkeypatch):
+    monkeypatch.setenv('NOEMA_LLM_MODEL', 'auto')
+    monkeypatch.setenv('NOEMA_LLM_API_URL', 'https://litellm.poinnetworks.net/v1/chat/completions')
+    monkeypatch.setenv('NOEMA_LLM_API_KEY', 'inert-test-value')
+    monkeypatch.setattr(gate, 'reject_private_llm_url', lambda _: None)
+    diff = 'diff --git a/a.py b/a.py\n--- a/a.py\n+++ b/a.py\n@@ -0,0 +1,1000 @@\n' + '+line\n'*1000
+    class Captured(BaseException):
+        pass
+    requests = []
+    def capture(request):
+        requests.append(request.data)
+        raise Captured()
+    with patch.object(gate.urllib.request, 'build_opener') as opener:
+        opener.return_value.open.side_effect = capture
+        with pytest.raises(Captured):
+            gate.call_llm('o/r', 1, {}, diff, False, 'a'*40)
+    payload = json.loads(requests[0])
+    prompt = payload['messages'][1]['content']
+    locations = json.loads(next(line.split(': ', 1)[1] for line in prompt.splitlines()
+                                if line.startswith('Allowed changed-side locations: ')))
+    assert not locations['truncated']
+    assert len(locations['locations']) == 1000
+    assert prompt.endswith(diff)
+    assert len(requests[0]) <= gate.MAX_AUTO_REVIEW_REQUEST_BYTES
+
+
+def test_auto_fetch_preserves_complete_diff(monkeypatch):
+    monkeypatch.setenv('NOEMA_LLM_MODEL', 'auto')
+    monkeypatch.setenv('NOEMA_LLM_API_URL', 'https://litellm.poinnetworks.net/v1/chat/completions')
+    diff = 'diff --git a/a.py b/a.py\n--- a/a.py\n+++ b/a.py\n@@ -0,0 +1,2 @@\n+' + 'a'*61000 + '\n+final_source_line\n'
+    monkeypatch.setattr(gate, 'run', lambda _: diff)
+    assert gate.fetch_diff('o/r', 1) == (diff, False)
+
+
+def test_real_serializer_rejects_oversized_auto_before_network(monkeypatch):
+    monkeypatch.setenv('NOEMA_LLM_MODEL', 'auto')
+    monkeypatch.setenv('NOEMA_LLM_API_URL', 'https://litellm.poinnetworks.net/v1/chat/completions')
+    monkeypatch.setenv('NOEMA_LLM_API_KEY', 'inert-test-value')
+    monkeypatch.setattr(gate, 'reject_private_llm_url', lambda _: None)
+    with patch.object(gate.urllib.request, 'build_opener') as opener:
+        with pytest.raises(RuntimeError, match='complete review input exceeds'):
+            gate.call_llm('o/r', 1, {'headRefOid':'a'*40}, '+'+'x'*500000, False, 'a'*40)
+        opener.assert_not_called()
+
+
+@pytest.mark.parametrize('model,url', [('auto','https://example.com/v1/chat/completions'), ('orchestrator/auto','https://litellm.poinnetworks.net/v1/chat/completions')])
+def test_other_routes_keep_existing_bound(monkeypatch, model, url):
+    monkeypatch.setenv('NOEMA_LLM_MODEL', model)
+    monkeypatch.setenv('NOEMA_LLM_API_URL', url)
+    monkeypatch.setattr(gate, 'run', lambda _: 'x'*61000)
+    text, truncated = gate.fetch_diff('o/r', 1)
+    assert truncated and len(text) <= gate.MAX_DIFF_CHARS
+
+
+def test_auto_rejects_partial_input(monkeypatch):
+    monkeypatch.setenv('NOEMA_LLM_MODEL', 'auto')
+    monkeypatch.setenv('NOEMA_LLM_API_URL', 'https://litellm.poinnetworks.net/v1/chat/completions')
+    monkeypatch.setenv('NOEMA_LLM_API_KEY', 'inert-test-value')
+    monkeypatch.setattr(gate, 'reject_private_llm_url', lambda _: None)
+    with patch.object(gate.urllib.request, 'build_opener') as opener:
+        with pytest.raises(RuntimeError, match='untruncated'):
+            gate.call_llm('o/r', 1, {}, 'partial', True, 'a'*40)
+        opener.assert_not_called()