feat(governance): 新增監控合約降噪矩陣

apps/api/src/api/v1/agents.py

@@ -65,6 +65,9 @@ from src.services.runtime_surface_inventory import (
 from src.services.gitea_workflow_runner_health import (
     load_latest_gitea_workflow_runner_health,
 )
+from src.services.observability_contract_matrix import (
+    load_latest_observability_contract_matrix,
+)
 from src.services.package_supply_chain_inventory import (
     load_latest_package_supply_chain_inventory,
 )
@@ -536,6 +539,34 @@ async def get_gitea_workflow_runner_health() -> dict[str, Any]:
         ) from exc
 
 
+@router.get(
+    "/observability-contract-matrix",
+    response_model=dict[str, Any],
+    summary="取得監控合約與降噪機會矩陣",
+    description=(
+        "讀取最新已提交的 Prometheus / Alertmanager / Grafana / SigNoz / ClickHouse / Sentry "
+        "只讀 observability matrix；此端點不修改 alert rules、不呼叫 silence API、"
+        "不建立 Grafana dashboard、不改 SigNoz / Sentry 設定、不讀 Secret payload、"
+        "不送 Telegram 測試通知、不觸發 monitoring deploy。"
+    ),
+)
+async def get_observability_contract_matrix() -> dict[str, Any]:
+    """Return the latest read-only observability contract matrix."""
+    try:
+        return await asyncio.to_thread(load_latest_observability_contract_matrix)
+    except FileNotFoundError as exc:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=str(exc),
+        ) from exc
+    except (json.JSONDecodeError, ValueError) as exc:
+        logger.error("observability_contract_matrix_invalid", error=str(exc))
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="監控合約與降噪機會矩陣快照無效",
+        ) from exc
+
+
 @router.get(
     "/backup-dr-target-inventory",
     response_model=dict[str, Any],

apps/api/src/services/observability_contract_matrix.py

@@ -0,0 +1,232 @@
+"""
+Observability contract and noise-reduction matrix snapshot.
+
+Loads the latest committed, read-only Prometheus / Alertmanager / SigNoz /
+Grafana observability contract matrix. This module never mutates alert rules,
+routes, receivers, silences, dashboards, webhooks, collectors, secrets,
+notifications, workflows, or runtime state.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+from src.services.snapshot_paths import default_evaluations_dir
+
+_DEFAULT_EVALUATIONS_DIR = default_evaluations_dir(Path(__file__))
+_SNAPSHOT_PATTERN = "observability_contract_matrix_*.json"
+_SCHEMA_VERSION = "observability_contract_matrix_v1"
+
+
+def load_latest_observability_contract_matrix(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed observability contract matrix snapshot."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no observability contract matrix snapshots found in {directory}")
+
+    latest = candidates[-1]
+    with latest.open(encoding="utf-8") as handle:
+        payload = json.load(handle)
+
+    if not isinstance(payload, dict):
+        raise ValueError(f"{latest}: expected JSON object")
+    _require_schema(payload, _SCHEMA_VERSION, str(latest))
+    _require_read_only_boundaries(payload, str(latest))
+    _require_operation_boundaries(payload, str(latest))
+    _require_rollup_consistency(payload, str(latest))
+    _require_surface_evidence(payload, str(latest))
+    _require_noise_opportunities(payload, str(latest))
+    _require_operator_denials(payload, str(latest))
+    _require_no_plaintext_secret_payload_keys(payload, str(latest))
+    return payload
+
+
+def _require_schema(payload: dict[str, Any], expected: str, label: str) -> None:
+    actual = payload.get("schema_version")
+    if actual != expected:
+        raise ValueError(f"{label}: expected schema_version={expected}, got {actual!r}")
+
+
+def _require_read_only_boundaries(payload: dict[str, Any], label: str) -> None:
+    program_status = payload.get("program_status") or {}
+    if program_status.get("read_only_mode") is not True:
+        raise ValueError(f"{label}: program_status.read_only_mode must be true")
+
+    approval_boundaries = payload.get("approval_boundaries") or {}
+    allowed = sorted(key for key, value in approval_boundaries.items() if value is not False)
+    if allowed:
+        raise ValueError(f"{label}: approval boundaries must remain false: {allowed}")
+
+
+def _require_operation_boundaries(payload: dict[str, Any], label: str) -> None:
+    boundaries = payload.get("operation_boundaries") or {}
+    if boundaries.get("read_only_api_allowed") is not True:
+        raise ValueError(f"{label}: read_only_api_allowed must be true")
+
+    blocked_flags = {
+        "prometheus_rule_write_allowed",
+        "prometheus_reload_allowed",
+        "alertmanager_route_write_allowed",
+        "alertmanager_receiver_change_allowed",
+        "alertmanager_to_openclaw_allowed",
+        "silence_create_allowed",
+        "grafana_dashboard_write_allowed",
+        "grafana_api_write_allowed",
+        "signoz_query_mutation_allowed",
+        "signoz_webhook_change_allowed",
+        "sentry_webhook_change_allowed",
+        "otel_collector_deploy_allowed",
+        "event_exporter_restart_allowed",
+        "secret_read_allowed",
+        "secret_plaintext_allowed",
+        "notification_send_allowed",
+        "external_api_call_allowed",
+        "live_prometheus_query_allowed",
+        "workflow_trigger_allowed",
+        "deploy_trigger_allowed",
+        "reload_trigger_allowed",
+        "runtime_execution_allowed",
+    }
+    allowed = sorted(flag for flag in blocked_flags if boundaries.get(flag) is not False)
+    if allowed:
+        raise ValueError(f"{label}: operation boundaries must remain false: {allowed}")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    surfaces = payload.get("observability_surfaces") or []
+    opportunities = payload.get("noise_reduction_opportunities") or []
+    gaps = payload.get("classification_gaps") or []
+    rollups = payload.get("rollups") or {}
+
+    if rollups.get("total_surfaces") != len(surfaces):
+        raise ValueError(f"{label}: rollups.total_surfaces must match observability_surfaces")
+    if rollups.get("by_kind") != _count_by(surfaces, "kind"):
+        raise ValueError(f"{label}: rollups.by_kind must match observability_surfaces")
+    if rollups.get("by_status") != _count_by(surfaces, "status"):
+        raise ValueError(f"{label}: rollups.by_status must match observability_surfaces")
+    if rollups.get("by_evidence_status") != _count_by(surfaces, "evidence_status"):
+        raise ValueError(f"{label}: rollups.by_evidence_status must match observability_surfaces")
+    if rollups.get("by_noise_policy_status") != _count_by(surfaces, "noise_policy_status"):
+        raise ValueError(f"{label}: rollups.by_noise_policy_status must match observability_surfaces")
+
+    action_required = sorted(
+        surface.get("surface_id")
+        for surface in surfaces
+        if surface.get("status") == "action_required"
+    )
+    if sorted(rollups.get("surface_ids_requiring_action") or []) != action_required:
+        raise ValueError(f"{label}: rollups.surface_ids_requiring_action must match surfaces")
+
+    proposal_only_surfaces = sorted(
+        surface.get("surface_id")
+        for surface in surfaces
+        if surface.get("noise_policy_status") == "proposal_only"
+    )
+    if sorted(rollups.get("surface_ids_with_proposal_only_noise_policy") or []) != proposal_only_surfaces:
+        raise ValueError(
+            f"{label}: rollups.surface_ids_with_proposal_only_noise_policy must match surfaces"
+        )
+
+    approval_required = sorted(
+        opportunity.get("opportunity_id")
+        for opportunity in opportunities
+        if opportunity.get("status") == "approval_required"
+    )
+    if rollups.get("noise_reduction_opportunities_total") != len(opportunities):
+        raise ValueError(f"{label}: rollups.noise_reduction_opportunities_total must match opportunities")
+    if sorted(rollups.get("approval_required_opportunity_ids") or []) != approval_required:
+        raise ValueError(f"{label}: rollups.approval_required_opportunity_ids must match opportunities")
+
+    if sorted(rollups.get("classification_gap_ids") or []) != sorted(gap.get("gap_id") for gap in gaps):
+        raise ValueError(f"{label}: rollups.classification_gap_ids must match classification_gaps")
+
+
+def _require_surface_evidence(payload: dict[str, Any], label: str) -> None:
+    surfaces = payload.get("observability_surfaces") or []
+    missing = sorted(
+        surface.get("surface_id")
+        for surface in surfaces
+        if not surface.get("coverage_contract")
+        or not surface.get("evidence_refs")
+        or not surface.get("next_action")
+    )
+    if missing:
+        raise ValueError(f"{label}: observability_surfaces must include contract, evidence, next_action: {missing}")
+
+
+def _require_noise_opportunities(payload: dict[str, Any], label: str) -> None:
+    opportunities = payload.get("noise_reduction_opportunities") or []
+    non_proposal = sorted(
+        opportunity.get("opportunity_id")
+        for opportunity in opportunities
+        if opportunity.get("proposal_only") is not True
+    )
+    if non_proposal:
+        raise ValueError(f"{label}: noise opportunities must stay proposal_only: {non_proposal}")
+
+    required_ids = {
+        "prometheus_noise_rule_tuning",
+        "alertmanager_grouping_inhibit_tuning",
+        "success_notification_quiet_policy",
+    }
+    present = {opportunity.get("opportunity_id") for opportunity in opportunities}
+    if not required_ids.issubset(present):
+        raise ValueError(f"{label}: missing required noise-reduction opportunities")
+
+
+def _require_operator_denials(payload: dict[str, Any], label: str) -> None:
+    contract = payload.get("operator_contract") or {}
+    must_not_interpret_as = set(contract.get("must_not_interpret_as") or [])
+    required_denials = {
+        "Prometheus alert rule 修改批准",
+        "Alertmanager receiver / route 修改批准",
+        "Alertmanager 指向 OpenClaw receiver 批准",
+        "Silence 建立或維護窗口批准",
+        "Grafana dashboard 寫入批准",
+        "SigNoz / Sentry webhook 設定修改批准",
+        "Secret 已讀取或可輸出",
+        "Telegram 測試通知批准",
+        "deploy / reload / workflow 觸發批准",
+        "runtime execution 授權",
+    }
+    if not required_denials.issubset(must_not_interpret_as):
+        raise ValueError(f"{label}: operator_contract.must_not_interpret_as is missing required denials")
+
+    route_policy = str(contract.get("alertmanager_route_policy") or "")
+    if "OpenClaw" not in route_policy or "不接收 Alertmanager webhook" not in route_policy:
+        raise ValueError(f"{label}: operator_contract.alertmanager_route_policy must block OpenClaw receiver use")
+
+
+def _require_no_plaintext_secret_payload_keys(value: Any, label: str, path: str = "$") -> None:
+    if isinstance(value, dict):
+        forbidden_key_fragments = {
+            "secret_value",
+            "token_value",
+            "authorization_header",
+            "private_key",
+            "webhook_secret",
+            "runner_token",
+            "signoz_token",
+            "sentry_dsn",
+        }
+        for key, nested in value.items():
+            lowered = str(key).lower()
+            if any(fragment in lowered for fragment in forbidden_key_fragments):
+                raise ValueError(f"{label}: forbidden secret payload key at {path}.{key}")
+            _require_no_plaintext_secret_payload_keys(nested, label, f"{path}.{key}")
+    elif isinstance(value, list):
+        for index, nested in enumerate(value):
+            _require_no_plaintext_secret_payload_keys(nested, label, f"{path}[{index}]")
+
+
+def _count_by(items: list[dict[str, Any]], key: str) -> dict[str, int]:
+    counts: dict[str, int] = {}
+    for item in items:
+        value = item.get(key)
+        counts[value] = counts.get(value, 0) + 1
+    return counts