AutoGineer
Generated banner for Cloud Monitoring Alert Configuration
Generated logo for Cloud Monitoring Alert Configuration

Cloud Monitoring Alert Configuration

GoogleOfficialPublished by Google

  • Skill
  • Workflow
  • Cloud and DevOps
  • Apache-2.0
  • Low risk
  • Has scripts
Download ZIP

From the author

Configures Cloud Monitoring alerting policies using Terraform and PromQL across Cloud Run, Google Kubernetes Engine (GKE), Compute Engine, AI Hypercomputer reservations, Vertex AI Agent Engine, and generic Google Cloud resources. Use when creating, reviewing, migrating, or deduplicating Terraform alerting policies for latency, traffic, error rates, saturation, host or reservation maintenance, GPU XID insights, or AI agent telemetry. Don't use for standalone PromQL query authoring without alerting, log querying, or billing cost audits.
— the description by Google

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

This skill provides a structured workflow for defining, reviewing, and managing Cloud Monitoring alerting policies. It leverages Terraform for infrastructure-as-code and PromQL for metric conditions, ensuring consistent observability across Google Cloud resources including Cloud Run, GKE, Compute Engine, and AI Agent platforms. The skill enforces best practices like the Four Golden Signals, dynamic grouping, and proper labeling, while providing clear paths for delegating SLO-specific configurations and notification channel management.

What it does

  • Generates parameterized Terraform configurations for Cloud Monitoring alert policies.
  • Implements PromQL-based conditions for high-signal monitoring.
  • Provides resource-specific guidance for GKE, Cloud Run, Compute Engine, and AI agents.
  • Enforces standardized labeling for all generated alerting resources.
  • Integrates with specialized scripts to analyze traffic patterns for AI agent telemetry.
  • Delegates SLO and notification channel configuration to dedicated skills.

When to use it

  • When creating or migrating Terraform-based alerting policies for Google Cloud infrastructure.
  • When setting up monitoring for AI agent reliability, quality, and cost telemetry.
  • When deduplicating existing alert policies to improve signal-to-noise ratios.
  • When configuring alerts for latency, traffic, error rates, or saturation across cloud services.

Advantages

  • Ensures consistent observability standards across diverse cloud environments.
  • Reduces manual configuration errors through structured Terraform templates.
  • Promotes best practices like dynamic grouping and the Four Golden Signals.
  • Maintains clear separation of concerns by delegating SLO and notification tasks.
  • Provides data-driven traffic analysis for accurate AI agent alert thresholds.

Usage examples

“Create a Terraform alerting policy for high latency on my GKE service.”
The agent will guide you through selecting the appropriate GKE PromQL queries, generate the google_monitoring_alert_policy Terraform resource, and prompt you to attach notification channels.
“Configure a cost alert for my Vertex AI Agent based on token usage.”
The agent will run the traffic analysis script to determine your workload pattern and generate a PromQL-based alert policy tailored to your agent's historical token consumption.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/google/cloud-monitoring-alert-configuration/download

  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip cloud-monitoring-alert-configuration.zip -d .agents/skills

Documentation by Google

Configure Cloud Monitoring Alerts

Configure high-signal alerting policies in Cloud Monitoring using Terraform (google_monitoring_alert_policy) and PromQL (condition_prometheus_query_language).

Usage Attribution

For all gcloud CLI, REST API, and MCP attribution requirements when executing commands with this skill, follow the instructions in Usage Attribution.

Enforce Core Alerting Rules

  1. Apply the Four Golden Signals: Cover latency, traffic, errors, and saturation for the target workload.

  2. Use Dynamic Grouping: Aggregate metrics by resource labels and parameterize thresholds with Terraform variables. Never hardcode ephemeral pod or instance identifiers.

  3. Default to PromQL: Use Prometheus Query Language (PromQL) for metric conditions, and use log-based alerts (condition_matched_log or google_logging_metric) for lifecycle and audit events.

  4. Attach Skill Attribution Labels: Always include user_labels with created-with-google-skill = "cloud-monitoring-alert-configuration" on every generated google_monitoring_alert_policy resource.

  5. Require User Confirmation Before Mutating Resources: Never execute commands that modify infrastructure, enable telemetry, create online monitors, or apply Terraform without waiting for explicit user confirmation.

  6. Prompt for Notification Channels: Alerting policies without notification channels incur charges without notifying anyone during an incident. Whenever notification channels are not specified in the user prompt across any workflow, explicitly inform the user that no notification channels are attached to the alerting policies, advise them to configure and attach notification channels before deploying, and ask at the end of your response if they want to configure notification channels.

Route by Target Resource Type

Identify the target resource type and read only the matching reference file directly. Never run directory listings or workspace file searches to look for queries or .tf files unless the prompt explicitly provides a file path. If a user asks to fix or validate a PromQL query without providing the query text or a file path, construct a representative broken-and-corrected query example directly.

Configure Cloud Run Alerts
Configure Google Kubernetes Engine Alerts
Configure AI Agent Alerts
Configure Compute Engine and AI Hypercomputer Alerts

  • Compute Engine VMs, MIGs, GPU XID Insights, Host or Reservation Maintenance (type=scheduled, unscheduled, or multiple), and Machine or Domain Health: compute_engine_alerts.md

Configure Generic Resource Alerts
Delegate Service Level Objective Alerts

Delegate to the google-cloud-slo-alert-configuration skill. That skill owns the SLO setup wizard, the burn-rate PromQL templates (references/promql_templates.md), the SLI metric mapping (references/service_metrics.md), and the App Hub scoping labels (references/app_hub_labels.md). Do NOT duplicate those templates here or claim that this skill owns its own SLO PromQL template library.

Delegate whenever the user asks for an availability or latency SLO, an error budget or burn-rate alert, or a Service Level Objective on any resource. When SLOs are only one part of a broader request (for example, a full production alerting suite for a Cloud Run service), delegate the SLO policies and continue generating the remaining golden-signal policies from the resource-specific reference in the same response.

Every resource reference carries a Context to hand off list naming the SLI metric, its good or bad filter, and the grouping labels — under a ## Delegate Service Level Objective Alerts heading for Cloud Run and generic resources, and inside the SLO section itself for GKE and Agent Platform. Read it and pass that context on, so the SLO skill does not repeat discovery you have already done. When delegating, you MUST also:

  1. State the SLO target and window explicitly if the user gave them, or let the SLO skill apply its SRE defaults (99.9% over a rolling 28 days) if they did not.

  2. Require the returned policies to carry user_labels with created-with-google-skill = "cloud-monitoring-alert-configuration" so attribution stays consistent with the rest of the generated suite. This literal value does NOT change when you delegate: the label records the entry-point skill, not whichever skill supplied the template. Do NOT set created-with-google-skill = "google-cloud-slo-alert-configuration" on policies generated through this delegation.

  3. Preserve the burn-rate duration contract, which google-cloud-slo-alert-configuration does not specify: Fast Burn (1h and 5m windows, factor 14.4) carries duration = "300s", and Slow Burn ([3d] and [6h] lookback, factor 1.0) omits the duration field entirely — do not emit duration = "0s" in its place. Cloud Monitoring rejects a non-zero retest buffer when the lookback window exceeds 25 hours.

  4. When no notification channels are provided, inform the user that no notification channels are attached to the alerting policies, advise them to configure and attach notification channels, and ask if they want to configure notification channels.

[!IMPORTANT] Always tell the user about the referral, even on the fallback path. Some reference files in this skill include standalone burn-rate templates to use when the google-cloud-slo-alert-configuration skill is not installed. If you emit policies from those fallback templates, your response MUST still tell the user that google-cloud-slo-alert-configuration owns SLO configuration, state that the emitted policies use the local fallback templates, and list the handoff context. Never silently present SLO policies as if this skill owned them.

[!NOTE] Exception: Agent Platform (OpenTelemetry-instrumented AI agents on any runtime, including Vertex AI Agent Engine and Cloud Run) is the one resource whose Fast and Slow burn policies live in this skill and remain authoritative. Delegate only for SLO customization — see Reliability.

Delegate Related Observability Tasks

  • Notification Channels: Delegate to cloud-monitoring-notification-channels-configuration to configure email, PagerDuty, Pub/Sub, or Slack channels. When an alerting policy has no notification channels attached, inform the user that no notification channels are attached and that they should configure notification channels and attach them to the alerting policy.

  • PromQL Authoring: Delegate to cloud-monitoring-promql-query for standalone PromQL query authoring or metric exploration.

Follow the Alert Configuration Workflow

  1. Discover: Identify target resources, metric schemas, and scoping projects using the lookup commands in the matching reference file (such as discover_metric_schema.py for Compute Engine metrics and other Google Cloud resource metrics, or gather_agent_info.py, check_telemetry.py, analyze_traffic.py, create_online_monitor.py, list_trace_scope_table_names.py, and list_log_scope_table_names.py for AI agents).

  2. Route: Load the reference file for the target resource type or delegate SLOs to google-cloud-slo-alert-configuration.

  3. Generate: Produce parameterized Terraform HCL configurations with required user_labels.

  4. Validate: When the loaded reference file requires a validation script (such as lint_promql.py, validate_config.py, lint_syntax.py, or scan_duplicates.py), run or print that exact script command. Do not list or search the scripts/ directory.

  5. Explain and Prompt: Explain each alert policy in plain English. When no notification channels are provided, tell the user that no notification channels are attached to the alerting policy, advise them to configure and attach notification channels, and ask if they want to configure notification channels.

Files

references/agent_platform_safety_alert_policies.md

Configure Safety Alert Policies

These alert policies help users detect high trigger rates of Model Armor policy violations in their deployed agents caused by prompt injections, false positives, or hallucinations.

Table of Contents

  • Verify Prerequisites: Lines 17-71

    • Retrieve the Trace Scope Observability Analytics Table: Lines 19-25

    • Link the BigQuery Dataset: Lines 26-71

  • Follow Critical Rules: Lines 72-194

    • Configure High Model Armor Safety Policy Trigger Rate: Lines 81-194

  • Run Tooling Scripts: Lines 195-202

  • Validate and Finalize: Lines 203-209

Verify Prerequisites

Retrieve the Trace Scope Observability Analytics Table

Retrieve the Observability Analytics table where all the Trace scope spans are stored. See the list_trace_scope_table_names entry in Run Tooling Scripts below for instructions on running the retrieval tool.

Link the BigQuery Dataset

A linked BigQuery dataset instance is required to configure alert policies depending on Observability Analytics trace data. This dataset MUST hold views for the _Trace trace bucket of the target Google Cloud project.

Applicable BigQuery datasets have a "type": "LINKED" attribute and reference the _Trace bucket in their description. To list all existing BigQuery datasets in a project, run the following command:

bq ls --format=prettyjson --project_id=$PROJECT_ID

Where:

  • PROJECT_ID: The target Google Cloud project ID.

If the Google Cloud project does not have a linked BigQuery dataset for the Trace bucket, you MUST EXPLICITLY ask the user for confirmation before proceeding to create one. If the user does not approve, you MUST skip creating safety alert policies. Otherwise, create a linked dataset by running the following command:

gcloud beta observability buckets datasets links create \
  projects/$PROJECT_ID/locations/$LOCATION/buckets/$BUCKET_ID/datasets/$DATASET_ID/links/$LINK_ID \
 --dataset=$DATASET_ID \
 --bucket=$BUCKET_ID \
 --location=$LOCATION \
 --project=$PROJECT_ID

Where:

  • PROJECT_ID: The target Google Cloud project ID.

  • LOCATION: The location of the observability buckets. If unknown, the location is included in the Observability Analytics table name retrieved in Retrieve the Trace Scope Observability Analytics Table.

  • BUCKET_ID: The ID of the observability bucket. By default, this ID is _Trace.

  • DATASET_ID: The ID of the dataset. By default, the trace data is stored in a dataset named Spans.

  • LINK_ID: The name of the new BigQuery dataset. Use trace_bq_linked_dataset as the default.

Follow Critical Rules

  • Alert policy configuration: You MUST configure the following Safety alerting policy. Proceed to the Validate and Finalize section below:

    1. High Model Armor Safety Policy Trigger Rate (15-Minute window)

  • SQL format: Configure all Safety alert policies using SQL queries. Do NOT use any other language such as PromQL or Monitoring Query Language (MQL).

Configure High Model Armor Safety Policy Trigger Rate
Define the Safety Telemetry Query

Use this query to configure the alert policy. Replace each variable enclosed in curly braces (for example, {placeholder}) with the corresponding values provided below:

WITH
  trace_to_agent AS (
  SELECT
    DISTINCT trace_id,
    JSON_VALUE(resource.attributes, '$."cloud.platform"') AS cloud_platform,
    JSON_VALUE(resource.attributes, '$."cloud.resource_id"') AS agent_id,
    JSON_VALUE(attributes, '$."gen_ai.agent.name"') AS agent_name
  FROM
    `{trace_scope_table_name}`
  WHERE
    JSON_VALUE(resource.attributes, '$."cloud.platform"') = "gcp.agent_engine"
      AND
    JSON_VALUE(resource.attributes, '$."cloud.resource_id"') IS NOT NULL
      AND
    JSON_VALUE(attributes, '$."gen_ai.agent.name"') IS NOT NULL
),
  all_model_armor_spans AS (
  SELECT
    s.*,
    t2a.agent_name,
    t2a.agent_id
  FROM
    `{trace_scope_table_name}` s
  FULL JOIN
    `trace_to_agent` t2a ON s.trace_id = t2a.trace_id
  WHERE
    JSON_VALUE(resource.attributes, '$."service.name"') = "modelarmor"
      AND
    name IN ('Request Path', 'Response Path')
      AND
    t2a.agent_name IS NOT NULL
      AND
    t2a.agent_id IS NOT NULL
)
SELECT
  agent_id,
  agent_name,
  violation,
  JSON_VALUE(attributes, '$."gen_ai.security.policy.id"') AS policy_id,
  COUNT(*) AS trigger_count,
  ROUND(COUNT(*) * 100.0 / SUM(COUNT(*)) OVER(PARTITION BY agent_id), 2) AS trigger_rate_percentage
FROM
  `all_model_armor_spans`
  LEFT JOIN
    UNNEST(JSON_VALUE_ARRAY(attributes, '$."gcp.modelarmor.violations"')) AS violation
GROUP BY
  agent_id,
  agent_name,
  violation,
  policy_id
QUALIFY
  violation IS NOT NULL
    AND
  trigger_rate_percentage >= 25
ORDER BY
  trigger_rate_percentage DESC

Where:

  • {trace_scope_table_name}: The Trace scope Observability Analytics table name retrieved in Verify Prerequisites.

Define Terraform HCL for Safety Alerts

Implement the alert policy resource using a condition_sql block with a row_count_test condition:

resource "google_monitoring_alert_policy" "high_model_armor_safety_policy_trigger_rate" {
  project      = var.project_id
  display_name = "Agent Safety - High Model Armor Safety Policy Trigger Rate"
  combiner     = "OR"

  conditions {
    display_name = "Agent High Model Armor Safety Policy Trigger Rate Exceeds 25%"
    condition_sql {
      query =  {model_armor_trigger_high_rate_telemetry_query}


      # Run evaluation periodically
      minutes {
        periodicity = 5
      }

      # Test triggers when one or more models violate the threshold (returning rows)
      row_count_test {
        comparison = "COMPARISON_GT"
        threshold  = 0
      }
    }
  }

  user_labels = {
    created-with-google-skill = "cloud-monitoring-alert-configuration"
  }
}

Where:

  • {model_armor_trigger_high_rate_telemetry_query}: The SQL alert query configured in Define the Safety Telemetry Query above.

Run Tooling Scripts

  1. list_trace_scope_table_names (Fallback): Use this script ONLY if gather_agent_info.py fails to retrieve the associated Trace scope Observability Analytics SQL table name:

    • Command: python3 scripts/agent_platform/list_trace_scope_table_names.py --project_id={gcp_project}

Validate and Finalize

  • Handle Script Failures: If list_trace_scope_table_names.py fails unexpectedly, verify the project ID. Ensure you have permissions to list trace scopes and linked datasets. If no table is found, verify whether a linked BigQuery dataset exists as described in Verify Prerequisites.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Python

Security

What our scanner noticed· Low risk

  • Scripts that your agent may run

This skill includes scripts that your agent may run. Read them before you use it.

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Published by
Google
Publisher
Google (Official)
Source
google/skills · skills/cloud/cloud-monitoring-alert-configuration · commit 2194d3b
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by Google.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download cloud-monitoring-alert-configuration.zip