diff --git a/apps/api/src/services/callback_action_spec.yaml b/apps/api/src/services/callback_action_spec.yaml
new file mode 100644
index 00000000..adbfa566
--- /dev/null
+++ b/apps/api/src/services/callback_action_spec.yaml
@@ -0,0 +1,430 @@
+# Telegram 分類按鈕 Action 規格註冊表
+# ============================================
+# Phase 5 Sprint 5.0 — 2026-04-14 Claude Sonnet 4.6
+# 相關: docs/superpowers/plans/2026-04-14-PHASE-5-category-buttons-completion.md
+#
+# 格式:
+# actions:
+#   <action_name>:
+#     label: <UI 顯示文字>
+#     emoji: <按鈕 emoji>
+#     risk: low | medium | high | critical
+#     callback_format: info | nonce   # info=2-part (查類), nonce=4-part (寫類)
+#     category: <alert_category>      # 綁定於 classify_alert_early 輸出
+#     mcp:
+#       provider: k8s | ssh | prometheus | signoz | database | internal
+#       tool: <MCP tool name>
+#       params:                        # 參數模板 (支援 {incident_id}/{labels.xxx}/{signals[0].xxx})
+#         <key>: <template>
+#     reply_format: text | code | url | truncated  # 執行結果 reply 樣式
+#     requires_multi_sig: false        # secops 類設 true
+#     timeout_sec: 10                  # MCP 呼叫 timeout
+#     description: <說明>
+
+version: "1.0"
+last_updated: "2026-04-14"
+
+actions:
+  # ==========================================================================
+  # 查類按鈕（無副作用，2-part info callback）
+  # ==========================================================================
+
+  check_process:
+    label: "查程序"
+    emoji: "🔍"
+    risk: low
+    callback_format: info
+    category: host_resource
+    mcp:
+      provider: ssh
+      tool: ssh_get_top_processes
+      params:
+        host: "{labels.instance}"
+        limit: 10
+    reply_format: code
+    timeout_sec: 8
+    description: "列出主機上 CPU 最高的 10 個程序"
+
+  check_log_nginx:
+    label: "查 Nginx Log"
+    emoji: "📋"
+    risk: low
+    callback_format: info
+    category: network
+    mcp:
+      provider: ssh
+      tool: ssh_get_nginx_error_log
+      params:
+        host: "{labels.instance}"
+        lines: 50
+    reply_format: truncated
+    timeout_sec: 10
+    description: "最後 50 行 Nginx error log"
+
+  check_log_container:
+    label: "查容器 Log"
+    emoji: "📋"
+    risk: low
+    callback_format: info
+    category: devops_tool
+    mcp:
+      provider: ssh
+      tool: ssh_get_container_logs
+      params:
+        host: "{labels.instance}"
+        container: "{labels.container}"
+        lines: 50
+    reply_format: truncated
+    timeout_sec: 10
+    description: "最後 50 行容器 stdout/stderr log"
+
+  check_log_minio:
+    label: "查 MinIO Log"
+    emoji: "📋"
+    risk: low
+    callback_format: info
+    category: storage
+    mcp:
+      provider: ssh
+      tool: ssh_get_container_logs
+      params:
+        host: "192.168.0.188"
+        container: "minio"
+        lines: 50
+    reply_format: truncated
+    timeout_sec: 10
+    description: "MinIO 容器最後 50 行 log"
+
+  check_port:
+    label: "查 Port"
+    emoji: "🔌"
+    risk: low
+    callback_format: info
+    category: network
+    mcp:
+      provider: ssh
+      tool: ssh_check_port
+      params:
+        host: "{labels.instance}"
+        port: "{labels.port}"
+    reply_format: text
+    timeout_sec: 5
+    description: "檢查主機上指定 port 是否 LISTEN"
+
+  check_health:
+    label: "查健康狀態"
+    emoji: "🔍"
+    risk: low
+    callback_format: info
+    category: external_site
+    mcp:
+      provider: prometheus
+      tool: prometheus_query
+      params:
+        query: 'probe_success{{instance="{labels.instance}"}}'
+    reply_format: text
+    timeout_sec: 5
+    description: "Blackbox probe 健康狀態"
+
+  check_pod_logs:
+    label: "查 Pod Log"
+    emoji: "📋"
+    risk: low
+    callback_format: info
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: k8s_get_pod_logs
+      params:
+        namespace: "{labels.namespace}"
+        pod: "{labels.pod}"
+        tail_lines: 50
+    reply_format: truncated
+    timeout_sec: 8
+    description: "Pod 最後 50 行 log"
+
+  describe_pod:
+    label: "詳細 Pod"
+    emoji: "📜"
+    risk: low
+    callback_format: info
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: k8s_describe_pod
+      params:
+        namespace: "{labels.namespace}"
+        pod: "{labels.pod}"
+    reply_format: truncated
+    timeout_sec: 5
+    description: "kubectl describe pod 完整資訊"
+
+  open_signoz:
+    label: "查 SignOz"
+    emoji: "🔍"
+    risk: low
+    callback_format: info
+    category: business
+    mcp:
+      provider: internal
+      tool: build_signoz_url
+      params:
+        service: "{labels.service}"
+    reply_format: url
+    timeout_sec: 1
+    description: "返回 SignOz deeplink"
+
+  open_flywheel:
+    label: "飛輪面板"
+    emoji: "📊"
+    risk: low
+    callback_format: info
+    category: flywheel_health
+    mcp:
+      provider: internal
+      tool: build_flywheel_url
+      params: {}
+    reply_format: url
+    timeout_sec: 1
+    description: "返回飛輪儀表板 URL"
+
+  # ==========================================================================
+  # 寫類按鈕（有副作用，4-part nonce callback）
+  # ==========================================================================
+
+  k8s_restart:
+    label: "重啟"
+    emoji: "🔄"
+    risk: medium
+    callback_format: nonce
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: kubectl_restart
+      params:
+        namespace: "{labels.namespace}"
+        deployment: "{labels.deployment}"
+    reply_format: text
+    timeout_sec: 30
+    description: "kubectl rollout restart deployment"
+
+  k8s_scale_up:
+    label: "擴容 +1"
+    emoji: "📈"
+    risk: medium
+    callback_format: nonce
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: kubectl_scale
+      params:
+        namespace: "{labels.namespace}"
+        deployment: "{labels.deployment}"
+        replicas_delta: 1
+    reply_format: text
+    timeout_sec: 10
+    description: "副本數 +1（單調遞增）"
+
+  k8s_scale_down:
+    label: "縮容 -1"
+    emoji: "📉"
+    risk: medium
+    callback_format: nonce
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: kubectl_scale
+      params:
+        namespace: "{labels.namespace}"
+        deployment: "{labels.deployment}"
+        replicas_delta: -1
+        min_replicas: 1  # 禁止 scale to 0
+    reply_format: text
+    timeout_sec: 10
+    description: "副本數 -1（禁止到 0）"
+
+  k8s_rollback:
+    label: "回滾"
+    emoji: "⏪"
+    risk: high
+    callback_format: nonce
+    category: kubernetes
+    mcp:
+      provider: k8s
+      tool: kubectl_rollout_undo  # 需 Sprint 5.3 新增此 MCP tool
+      params:
+        namespace: "{labels.namespace}"
+        deployment: "{labels.deployment}"
+    reply_format: text
+    timeout_sec: 60
+    description: "kubectl rollout undo（回到上一版）"
+
+  host_restart_service:
+    label: "重啟服務"
+    emoji: "🔄"
+    risk: high
+    callback_format: nonce
+    category: host_resource
+    mcp:
+      provider: ssh
+      tool: ssh_systemctl_restart
+      params:
+        host: "{labels.instance}"
+        service: "{labels.service}"
+    reply_format: text
+    timeout_sec: 30
+    description: "systemctl restart 服務（主機層）"
+
+  host_clear_log:
+    label: "清 Log"
+    emoji: "🗑"
+    risk: low
+    callback_format: nonce
+    category: host_resource
+    mcp:
+      provider: ssh
+      tool: ssh_clear_docker_logs
+      params:
+        host: "{labels.instance}"
+        container: "{labels.container}"
+    reply_format: text
+    timeout_sec: 10
+    description: "清空容器 log file"
+
+  docker_restart:
+    label: "重啟容器"
+    emoji: "🔄"
+    risk: high
+    callback_format: nonce
+    category: devops_tool
+    mcp:
+      provider: ssh
+      tool: ssh_docker_restart
+      params:
+        host: "{labels.instance}"
+        container: "{labels.container}"
+    reply_format: text
+    timeout_sec: 30
+    description: "docker restart <container>"
+
+  minio_restart:
+    label: "重啟 MinIO"
+    emoji: "🔄"
+    risk: high
+    callback_format: nonce
+    category: storage
+    mcp:
+      provider: ssh
+      tool: ssh_docker_restart
+      params:
+        host: "192.168.0.188"
+        container: "minio"
+    reply_format: text
+    timeout_sec: 30
+    description: "重啟 MinIO Docker 容器"
+
+  reload_nginx:
+    label: "重載 Nginx"
+    emoji: "🔄"
+    risk: low
+    callback_format: nonce
+    category: network
+    mcp:
+      provider: ssh
+      tool: ssh_reload_nginx
+      params:
+        host: "{labels.instance}"
+    reply_format: text
+    timeout_sec: 10
+    description: "nginx -s reload（不停機）"
+
+  renew_cert:
+    label: "更新憑證"
+    emoji: "🔐"
+    risk: medium
+    callback_format: nonce
+    category: ssl_cert
+    mcp:
+      provider: ssh
+      tool: ssh_renew_ssl
+      params:
+        host: "{labels.instance}"
+        domain: "{labels.domain}"
+    reply_format: text
+    timeout_sec: 60
+    description: "certbot renew"
+
+  # ==========================================================================
+  # 資安按鈕（CRITICAL，Multi-Sig 必要）
+  # ==========================================================================
+
+  secops_isolate:
+    label: "隔離資源"
+    emoji: "🚫"
+    risk: critical
+    callback_format: nonce
+    category: secops
+    requires_multi_sig: true
+    mcp:
+      provider: k8s
+      tool: kubectl_exec  # 需 Sprint 5.4 實作 NetworkPolicy apply
+      params:
+        namespace: "{labels.namespace}"
+        command: "apply_network_policy_isolate"
+        target_pod: "{labels.pod}"
+    reply_format: text
+    timeout_sec: 30
+    description: "套用 NetworkPolicy 阻斷所有流量進入 Pod"
+
+  secops_block_ip:
+    label: "封鎖來源 IP"
+    emoji: "⛔"
+    risk: critical
+    callback_format: nonce
+    category: secops
+    requires_multi_sig: true
+    mcp:
+      provider: ssh
+      tool: ssh_systemctl_restart  # 需擴充加 iptables rule
+      params:
+        host: "{labels.instance}"
+        service: "iptables-block"
+        source_ip: "{labels.attacker_ip}"
+    reply_format: text
+    timeout_sec: 30
+    description: "iptables DROP 來源 IP"
+
+  secops_evict:
+    label: "強制驅逐 Pod"
+    emoji: "🔄"
+    risk: critical
+    callback_format: nonce
+    category: secops
+    requires_multi_sig: true
+    mcp:
+      provider: k8s
+      tool: kubectl_delete
+      params:
+        namespace: "{labels.namespace}"
+        resource_type: pod
+        name: "{labels.pod}"
+    reply_format: text
+    timeout_sec: 30
+    description: "kubectl delete pod（會自動重建）"
+
+  secops_authorize:
+    label: "確認授權"
+    emoji: "✅"
+    risk: high
+    callback_format: nonce
+    category: secops
+    requires_multi_sig: false  # 確認授權不需 multi-sig
+    mcp:
+      provider: internal
+      tool: record_authorization
+      params:
+        user_id: "{callback.user_id}"
+        source: "{labels.instance}"
+    reply_format: text
+    timeout_sec: 5
+    description: "記錄授權，後續同 source 不再告警（24h）"
diff --git a/apps/api/src/services/callback_dispatcher.py b/apps/api/src/services/callback_dispatcher.py
new file mode 100644
index 00000000..847ab8c4
--- /dev/null
+++ b/apps/api/src/services/callback_dispatcher.py
@@ -0,0 +1,324 @@
+"""
+Telegram Callback Dispatcher — 分類按鈕統一調度
+================================================
+Phase 5 Sprint 5.0-5.1 — 2026-04-14 Claude Sonnet 4.6
+相關: docs/superpowers/plans/2026-04-14-PHASE-5-category-buttons-completion.md
+       ADR-079 分類按鈕完整化
+
+職責:
+1. 從 callback_action_spec.yaml 載入 action registry
+2. 接收 Telegram callback_data (action:incident_id or action:id:ts:rand)
+3. 驗證 nonce（寫類按鈕）或 allow info（查類按鈕）
+4. 依 spec 呼叫對應 MCP tool
+5. Reply 執行結果到原告警卡片（reply_to_message_id）
+
+設計原則:
+- Registry pattern — 新增按鈕只需 yaml 一行，無需改 dispatcher code
+- 模板變數: {incident_id} / {labels.xxx} / {signals[0].xxx} / {callback.user_id}
+- 所有 action 都有 audit log（寫類額外 nonce 驗證 log）
+- reply_to 原告警 message_id（從 Redis tg_msg:{incident_id}）
+
+遵守「禁止 Mock 測試鐵律」: 純邏輯 + MCP dispatch，測試用真實 registry。
+"""
+
+from __future__ import annotations
+
+import time
+from dataclasses import dataclass
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+import structlog
+import yaml
+
+logger = structlog.get_logger(__name__)
+
+
+# =============================================================================
+# Data Types
+# =============================================================================
+
+
+@dataclass
+class ActionSpec:
+    """從 callback_action_spec.yaml 載入的單一 action 規格"""
+
+    name: str
+    label: str
+    emoji: str
+    risk: str  # low | medium | high | critical
+    callback_format: str  # info | nonce
+    category: str
+    mcp_provider: str  # k8s | ssh | prometheus | signoz | database | internal
+    mcp_tool: str
+    mcp_params: dict[str, Any]
+    reply_format: str  # text | code | url | truncated
+    timeout_sec: int
+    description: str
+    requires_multi_sig: bool = False
+
+
+@dataclass
+class DispatchResult:
+    """Dispatcher 執行結果"""
+
+    success: bool
+    action: str
+    incident_id: str
+    user_id: int | None
+    result_text: str
+    error: str | None = None
+    duration_ms: float = 0.0
+
+
+# =============================================================================
+# Spec Registry
+# =============================================================================
+
+
+@lru_cache(maxsize=1)
+def load_action_registry() -> dict[str, ActionSpec]:
+    """
+    載入 callback_action_spec.yaml 並快取（進程內不重載，重啟 Pod 才更新）
+
+    Returns:
+        {action_name: ActionSpec}
+    """
+    spec_path = Path(__file__).parent / "callback_action_spec.yaml"
+    if not spec_path.exists():
+        logger.warning("callback_action_spec_not_found", path=str(spec_path))
+        return {}
+
+    with spec_path.open("r", encoding="utf-8") as f:
+        data = yaml.safe_load(f)
+
+    registry: dict[str, ActionSpec] = {}
+    for name, spec_dict in (data.get("actions") or {}).items():
+        mcp = spec_dict.get("mcp", {}) or {}
+        registry[name] = ActionSpec(
+            name=name,
+            label=spec_dict.get("label", name),
+            emoji=spec_dict.get("emoji", ""),
+            risk=spec_dict.get("risk", "medium"),
+            callback_format=spec_dict.get("callback_format", "info"),
+            category=spec_dict.get("category", ""),
+            mcp_provider=mcp.get("provider", ""),
+            mcp_tool=mcp.get("tool", ""),
+            mcp_params=mcp.get("params") or {},
+            reply_format=spec_dict.get("reply_format", "text"),
+            timeout_sec=int(spec_dict.get("timeout_sec", 10)),
+            description=spec_dict.get("description", ""),
+            requires_multi_sig=bool(spec_dict.get("requires_multi_sig", False)),
+        )
+    logger.info("callback_action_registry_loaded", count=len(registry))
+    return registry
+
+
+def get_action_spec(action_name: str) -> ActionSpec | None:
+    """查找單一 action 規格"""
+    return load_action_registry().get(action_name)
+
+
+def list_actions_for_category(alert_category: str) -> list[ActionSpec]:
+    """列出特定分類的所有可用 action（供 _build_inline_keyboard 使用）"""
+    return [
+        spec for spec in load_action_registry().values()
+        if spec.category == alert_category
+    ]
+
+
+# =============================================================================
+# Template Variable Substitution
+# =============================================================================
+
+
+def _resolve_template(template: Any, context: dict) -> Any:
+    """
+    遞迴替換模板變數。
+
+    支援:
+    - {incident_id}
+    - {labels.xxx} / {labels.xxx.yyy}
+    - {signals[0].xxx}
+    - {callback.user_id}
+
+    Example:
+        template = {"host": "{labels.instance}", "lines": 50}
+        context = {"labels": {"instance": "192.168.0.110"}, "incident_id": "INC-123"}
+        → {"host": "192.168.0.110", "lines": 50}
+    """
+    if isinstance(template, dict):
+        return {k: _resolve_template(v, context) for k, v in template.items()}
+    if isinstance(template, list):
+        return [_resolve_template(v, context) for v in template]
+    if isinstance(template, str) and "{" in template:
+        # 找出所有 {xxx} placeholder 並替換
+        import re
+        def _repl(m: re.Match) -> str:
+            key = m.group(1)
+            val = _lookup_context(key, context)
+            return str(val) if val is not None else m.group(0)
+        return re.sub(r"\{([a-zA-Z0-9_.\[\]]+)\}", _repl, template)
+    return template
+
+
+def _lookup_context(key: str, context: dict) -> Any:
+    """
+    從 context 查表（支援巢狀 key: labels.instance / signals[0].alert_name）
+    """
+    parts = key.replace("[", ".").replace("]", "").split(".")
+    cur: Any = context
+    for part in parts:
+        if part == "":
+            continue
+        if isinstance(cur, dict):
+            cur = cur.get(part)
+        elif isinstance(cur, list):
+            try:
+                cur = cur[int(part)]
+            except (ValueError, IndexError):
+                return None
+        else:
+            return None
+        if cur is None:
+            return None
+    return cur
+
+
+# =============================================================================
+# Dispatcher (Sprint 5.1)
+# =============================================================================
+
+
+async def dispatch_action(
+    action_name: str,
+    incident_id: str,
+    user_id: int | None = None,
+    labels: dict | None = None,
+    extra_context: dict | None = None,
+) -> DispatchResult:
+    """
+    執行 callback action — 依 spec 呼叫 MCP tool
+
+    Args:
+        action_name: action 名稱（對應 spec registry）
+        incident_id: 關聯 incident
+        user_id: Telegram user id（callback 來源）
+        labels: alert labels（供模板替換）
+        extra_context: 額外上下文（signals 等）
+
+    Returns:
+        DispatchResult（包含 result_text 供 reply 使用）
+    """
+    start = time.perf_counter()
+    spec = get_action_spec(action_name)
+    if not spec:
+        logger.warning("dispatch_action_unknown", action=action_name)
+        return DispatchResult(
+            success=False,
+            action=action_name,
+            incident_id=incident_id,
+            user_id=user_id,
+            result_text="",
+            error=f"Unknown action: {action_name}",
+            duration_ms=(time.perf_counter() - start) * 1000,
+        )
+
+    # 建立模板 context
+    context = {
+        "incident_id": incident_id,
+        "labels": labels or {},
+        "callback": {"user_id": user_id or 0},
+        **(extra_context or {}),
+    }
+    resolved_params = _resolve_template(spec.mcp_params, context)
+
+    # Audit log (all actions)
+    logger.info(
+        "dispatch_action_start",
+        action=action_name,
+        incident_id=incident_id,
+        user_id=user_id,
+        risk=spec.risk,
+        provider=spec.mcp_provider,
+        tool=spec.mcp_tool,
+        params=resolved_params,
+    )
+
+    # MCP 呼叫（Sprint 5.1 只實作骨架，實際 MCP dispatch 在 Sprint 5.2/5.3）
+    try:
+        # Sprint 5.1 TODO: 接入 MCP registry
+        # from src.plugins.mcp.registry import get_provider
+        # provider = get_provider(spec.mcp_provider)
+        # import asyncio
+        # mcp_result = await asyncio.wait_for(
+        #     provider.execute(spec.mcp_tool, resolved_params),
+        #     timeout=spec.timeout_sec,
+        # )
+        # result_text = _format_reply(mcp_result, spec.reply_format, spec.label, spec.emoji)
+
+        # Sprint 5.0 骨架：僅返回「尚未實作」提示
+        result_text = (
+            f"{spec.emoji} <b>{spec.label}</b>\n"
+            f"⚠️ Phase 5 Sprint 5.2+ 實作中（spec 已載入：{spec.mcp_provider}/{spec.mcp_tool}）\n"
+            f"參數: <code>{resolved_params}</code>"
+        )
+        duration = (time.perf_counter() - start) * 1000
+        logger.info(
+            "dispatch_action_stub",
+            action=action_name,
+            incident_id=incident_id,
+            duration_ms=round(duration, 1),
+        )
+        return DispatchResult(
+            success=True,
+            action=action_name,
+            incident_id=incident_id,
+            user_id=user_id,
+            result_text=result_text,
+            duration_ms=duration,
+        )
+    except Exception as e:
+        duration = (time.perf_counter() - start) * 1000
+        logger.error(
+            "dispatch_action_failed",
+            action=action_name,
+            incident_id=incident_id,
+            error=str(e),
+            duration_ms=round(duration, 1),
+        )
+        return DispatchResult(
+            success=False,
+            action=action_name,
+            incident_id=incident_id,
+            user_id=user_id,
+            result_text=f"{spec.emoji} {spec.label} 執行失敗",
+            error=str(e),
+            duration_ms=duration,
+        )
+
+
+def _format_reply(
+    mcp_result: Any, reply_format: str, label: str, emoji: str
+) -> str:
+    """
+    依 spec 格式化 reply 文字。
+
+    reply_format:
+    - text: 單行文字
+    - code: <code>...</code>
+    - truncated: 截斷到 500 字
+    - url: 直接返回 URL
+    """
+    header = f"{emoji} <b>{label}</b>"
+    if reply_format == "url":
+        return f"{header}\n{mcp_result}"
+    if reply_format == "code":
+        return f"{header}\n<code>{str(mcp_result)[:800]}</code>"
+    if reply_format == "truncated":
+        text = str(mcp_result)[:500]
+        if len(str(mcp_result)) > 500:
+            text += "...\n<i>(已截斷)</i>"
+        return f"{header}\n<pre>{text}</pre>"
+    return f"{header}\n{mcp_result}"
diff --git a/apps/api/tests/test_callback_dispatcher.py b/apps/api/tests/test_callback_dispatcher.py
new file mode 100644
index 00000000..c5728271
--- /dev/null
+++ b/apps/api/tests/test_callback_dispatcher.py
@@ -0,0 +1,192 @@
+"""
+Phase 5 Sprint 5.0-5.1 Callback Dispatcher 單元測試
+======================================================
+建立: 2026-04-14 台北深夜 Claude Sonnet 4.6
+
+覆蓋:
+- callback_action_spec.yaml 載入正確性
+- 24 個 action 都能解析
+- 模板變數替換（_resolve_template）
+- Context lookup（labels.instance / signals[0].alert_name）
+- dispatch_action 骨架（Sprint 5.0 階段返回 stub）
+
+🔴 遵循「禁止 Mock 測試鐵律」: 用真實 spec registry，不 mock。
+"""
+
+import pytest
+
+from src.services.callback_dispatcher import (
+    dispatch_action,
+    get_action_spec,
+    list_actions_for_category,
+    load_action_registry,
+    _lookup_context,
+    _resolve_template,
+)
+
+
+# =============================================================================
+# Registry loading
+# =============================================================================
+
+
+class TestRegistryLoading:
+    def test_registry_loads_all_24_actions(self):
+        registry = load_action_registry()
+        # 10 查類 + 10 寫類 + 4 secops = 24
+        assert len(registry) >= 20, f"expected >= 20 actions, got {len(registry)}"
+
+    def test_all_actions_have_required_fields(self):
+        registry = load_action_registry()
+        for name, spec in registry.items():
+            assert spec.name == name
+            assert spec.label
+            assert spec.risk in ("low", "medium", "high", "critical")
+            assert spec.callback_format in ("info", "nonce")
+            assert spec.mcp_provider, f"{name} missing mcp_provider"
+            assert spec.mcp_tool, f"{name} missing mcp_tool"
+
+    def test_secops_requires_multi_sig(self):
+        for sa in ("secops_isolate", "secops_block_ip", "secops_evict"):
+            spec = get_action_spec(sa)
+            assert spec and spec.requires_multi_sig is True, \
+                f"{sa} should require multi_sig"
+
+    def test_info_actions_dont_need_multi_sig(self):
+        spec = get_action_spec("check_process")
+        assert spec and spec.requires_multi_sig is False
+
+    def test_write_actions_use_nonce_format(self):
+        for wa in ("k8s_restart", "k8s_scale_up", "k8s_rollback", "host_restart_service"):
+            spec = get_action_spec(wa)
+            assert spec and spec.callback_format == "nonce", \
+                f"{wa} should use nonce format"
+
+    def test_query_actions_use_info_format(self):
+        for qa in ("check_process", "check_port", "open_signoz"):
+            spec = get_action_spec(qa)
+            assert spec and spec.callback_format == "info", \
+                f"{qa} should use info format"
+
+
+# =============================================================================
+# Category filtering
+# =============================================================================
+
+
+class TestCategoryFiltering:
+    def test_kubernetes_has_4_write_actions(self):
+        actions = list_actions_for_category("kubernetes")
+        write_actions = [a for a in actions if a.callback_format == "nonce"]
+        assert len(write_actions) >= 4, \
+            f"kubernetes should have at least 4 write actions, got {len(write_actions)}"
+
+    def test_secops_has_4_actions(self):
+        actions = list_actions_for_category("secops")
+        assert len(actions) == 4, f"secops should have 4 actions, got {len(actions)}"
+
+    def test_host_resource_has_mix(self):
+        actions = list_actions_for_category("host_resource")
+        assert len(actions) >= 2
+        assert any(a.callback_format == "info" for a in actions), "需至少 1 個查類"
+        assert any(a.callback_format == "nonce" for a in actions), "需至少 1 個寫類"
+
+
+# =============================================================================
+# Template variable resolution
+# =============================================================================
+
+
+class TestTemplateResolution:
+    def test_lookup_simple_key(self):
+        ctx = {"incident_id": "INC-123"}
+        assert _lookup_context("incident_id", ctx) == "INC-123"
+
+    def test_lookup_nested_labels(self):
+        ctx = {"labels": {"instance": "192.168.0.110"}}
+        assert _lookup_context("labels.instance", ctx) == "192.168.0.110"
+
+    def test_lookup_deep_nested(self):
+        ctx = {"labels": {"k8s": {"pod": "api-1"}}}
+        assert _lookup_context("labels.k8s.pod", ctx) == "api-1"
+
+    def test_lookup_list_index(self):
+        ctx = {"signals": [{"alert_name": "KubePodCrashLooping"}]}
+        assert _lookup_context("signals[0].alert_name", ctx) == "KubePodCrashLooping"
+
+    def test_lookup_missing_returns_none(self):
+        ctx = {"labels": {}}
+        assert _lookup_context("labels.instance", ctx) is None
+
+    def test_resolve_template_dict(self):
+        tpl = {"host": "{labels.instance}", "lines": 50}
+        ctx = {"labels": {"instance": "10.0.0.1"}}
+        out = _resolve_template(tpl, ctx)
+        assert out == {"host": "10.0.0.1", "lines": 50}
+
+    def test_resolve_keeps_unresolved(self):
+        """若 context 缺 key，保留原 {...}（便於 debug）"""
+        tpl = {"host": "{labels.missing}"}
+        ctx = {"labels": {}}
+        out = _resolve_template(tpl, ctx)
+        assert out == {"host": "{labels.missing}"}
+
+    def test_resolve_string_with_multiple_placeholders(self):
+        tpl = "host={labels.instance} port={labels.port}"
+        ctx = {"labels": {"instance": "10.0.0.1", "port": "9100"}}
+        out = _resolve_template(tpl, ctx)
+        assert out == "host=10.0.0.1 port=9100"
+
+
+# =============================================================================
+# dispatch_action stub (Sprint 5.0 骨架)
+# =============================================================================
+
+
+@pytest.mark.asyncio
+class TestDispatchActionStub:
+    async def test_unknown_action_returns_failure(self):
+        result = await dispatch_action(
+            action_name="unknown_action",
+            incident_id="INC-TEST-001",
+        )
+        assert result.success is False
+        assert "Unknown action" in (result.error or "")
+
+    async def test_check_process_stub_returns_spec_hint(self):
+        result = await dispatch_action(
+            action_name="check_process",
+            incident_id="INC-TEST-002",
+            user_id=12345,
+            labels={"instance": "192.168.0.110"},
+        )
+        assert result.success is True
+        # Sprint 5.0 階段 stub：應含「Sprint 5.2+ 實作中」提示 + 解析後的參數
+        assert "Sprint 5.2+" in result.result_text
+        assert "192.168.0.110" in result.result_text  # labels.instance 正確替換
+
+    async def test_k8s_restart_stub_with_labels(self):
+        result = await dispatch_action(
+            action_name="k8s_restart",
+            incident_id="INC-TEST-003",
+            labels={"namespace": "awoooi-prod", "deployment": "awoooi-api"},
+        )
+        assert result.success is True
+        # 驗證 namespace 和 deployment 都被替換
+        assert "awoooi-prod" in result.result_text
+        assert "awoooi-api" in result.result_text
+
+    async def test_dispatch_includes_duration(self):
+        result = await dispatch_action(
+            action_name="check_process",
+            incident_id="INC-TEST-004",
+            labels={"instance": "10.0.0.1"},
+        )
+        assert result.duration_ms >= 0
+        assert result.duration_ms < 5000  # stub 應極快
+
+    async def test_secops_action_flag_preserved(self):
+        """secops 動作在 dispatcher 結果中能識別 (供上層 Multi-Sig 處理)"""
+        spec = get_action_spec("secops_isolate")
+        assert spec and spec.requires_multi_sig is True
+        # dispatcher 本身不做 multi-sig 攔截（留給 callback_handler），只記錄 spec