OpenClaw Setup (one-time)#

This skill calls MCP tools prefixed with mcp__amg__*, so OpenClaw must have an MCP server registered under the exact name amg. Run this once per workspace before invoking the skill:

bash

openclaw mcp set amg '{"url":"https://<your-grafana-instance>/api/azure-mcp","transport":"streamable-http","headers":{"Authorization":"Bearer <your-token>"}}'

Replace <your-grafana-instance> with your Azure Managed Grafana endpoint and <your-token> with a valid Grafana service-account token (starts with glsa_). The server name must be amg — the skill's allowed-tools reference mcp__amg__* and will not find tools under any other name.

Verify the server is registered:

bash

openclaw mcp list

Official skill source: https://github.com/Azure/amg-skills

Runtime Context#

• Current UTC time: ! date -u +%Y-%m-%dT%H:%M:%SZ
• Config: ! cat memory/amg-check-storage-account/config.md 2>/dev/null || echo "NOT_CONFIGURED"
• Prior report: ! [ -f memory/amg-check-storage-account/report.md ] && echo "exists ($(grep -c '^### SA-' memory/amg-check-storage-account/report.md) bugs documented)" || echo "not found"
• Arguments: time-range=$0, subscription-override=$1

Known Issues: Before presenting findings, cross-reference results against memory/amg-check-storage-account/report.md.

Azure Storage Account Health Check#

Analyze Azure Storage Account health using a two-phase approach: a single amgmcp_pulse_check call for fleet-wide summary, followed by targeted deep dives into the top 7 most interesting accounts only.

Critical Constraints#

• Do NOT use subagents (Agent tool) for MCP queries. Subagents do not have access to MCP tools. All MCP tool calls must be made directly from the main conversation context.
• Deep dive limit: at most 7 accounts. Select the most interesting accounts from pulse check results. Do not deep-dive the entire fleet.
• Time format: ISO 8601 UTC with explicit from/ to — NEVER use timespan (it causes errors).
• UsedCapacity only supports interval PT1H — do NOT use PT6H or P1D for this metric.
• Parallelism cap: 30 concurrent MCP calls per batch. Reduce to 4-5 if rate-limited.

Prerequisites#

• An AMG-MCP server must be connected (the allowed-tools frontmatter references the MCP server name — update it if your server has a different name)
• The MCP server's Grafana service account token environment variable must be set

Configuration#

If Config shows NOT_CONFIGURED: Run First-Run Setup at the bottom of this file, then return here.

If Config is populated: Extract the datasource UID and subscription ID(s) from the pre-loaded Runtime Context above and use them for all queries. Use $1 as the subscription override if provided.

• Datasource UID: from ## Azure Monitor Datasource > UID
• Subscription ID(s): from ## Subscriptions (or $1 if provided)
• Resource Type: microsoft.storage/storageaccounts (lowercase)
• ARM ID template: /subscriptions/{SUB}/resourceGroups/{RG}/providers/Microsoft.Storage/storageAccounts/{name}

Time Range#

Default is 7 days ( pastDays: 7) for pulse check and deep-dive metrics, 24 hours for resource logs. If the user specifies a different range via $ARGUMENTS[0] (e.g., /amg-check-storage-account 3d), adjust accordingly. For resource log queries, keep the range narrow (1-2 days) to avoid timeouts.

---

Workflow#

Phase 1: Validate Datasource & Discover Accounts#

Step 1a: Validate Datasource

Call amgmcp_datasource_list with no parameters.

Search the results for a datasource with type equal to grafana-azure-monitor-datasource. Extract its uid.

• If it matches the configured UID, proceed.
• If it differs, update memory/amg-check-storage-account/config.md, warn the user, and use the new UID.
• If no Azure Monitor datasource is found, abort with a clear error.

Step 1b: Discover All Storage Accounts

Call amgmcp_query_resource_graph once using the configured datasource UID and subscription ID(s):

text

azureMonitorDatasourceUid: {DATASOURCE_UID}
query: |
  resources
  | where type == 'microsoft.storage/storageaccounts'
  | where subscriptionId in ({SUBSCRIPTION_IDS})
  | project name, resourceGroup, location, subscriptionId, properties.provisioningState
  | order by location asc, name asc

Replace {SUBSCRIPTION_IDS} with the configured subscription IDs formatted as comma-separated quoted strings (e.g., 'sub-id-1', 'sub-id-2').

Constructing the ARM resource ID: Use subscriptionId from each row:

text

/subscriptions/{subscriptionId}/resourceGroups/{resourceGroup}/providers/Microsoft.Storage/storageAccounts/{name}

Region summary: Derive from the account list by counting accounts per unique location value.

Note any accounts not in "Succeeded" provisioning state — flag them immediately.

If zero accounts are found, report "No Storage Accounts found" and stop.

Phase 2: Fleet-Wide Pulse Check#

Call amgmcp_pulse_check once to get a summary across all storage accounts:

text

azureMonitorDatasourceUid: {DATASOURCE_UID}
pastDays: 7
scenarios: storage_summary

If $1 provides a subscription ID, add subscriptionId to scope the scan. Otherwise, if the config has a single subscription, pass it.

After the pulse check, verify:

1. The number of scanned resources is close to the Phase 1 account count.
2. The scenario shows status: "completed".
3. If errors occurred, retry once. If still failing, note the failure in the report.

Cross-reference pulse check results with Phase 1 inventory to enrich each account with its resource group and region from the Resource Graph data.

Phase 3: Top 7 Deep Dive#

From the pulse check results, select at most 7 accounts for detailed investigation. Prioritize accounts with the most interesting signals:

1. Availability drops — any account below 99.9% availability
2. Highest error transaction counts — accounts with the most non-Success responses
3. Highest latency — accounts with the highest average E2E latency
4. Unusual capacity or traffic patterns — sudden capacity jumps, abnormally high ingress/egress
5. Diversity — prefer selecting accounts from different regions to maximize coverage

If the pulse check shows fewer than 7 accounts with notable signals, only deep-dive those that have something worth investigating. Do not pad to 7.

If the pulse check shows the entire fleet is healthy with no notable signals, skip Phase 3 entirely and report the fleet as healthy.

Step 3a: Deep Metrics

For each selected account, query these metrics in parallel using amgmcp_query_resource_metric. Compute from (matching pastDays from Phase 2) and to (now) in ISO 8601 UTC.

Metric NameAggregationIntervalPurposeAvailability``AveragePT6HAvailability trendSuccessE2ELatency``AveragePT6HClient-perceived latencySuccessE2ELatency``MaximumPT6HTail latency spikesSuccessServerLatency``AveragePT6HStorage backend latencyTransactions``TotalPT6HTotal volume (no filter)Transactions (errors)TotalPT6HError count ( filter: ResponseType ne 'Success')UsedCapacity``AveragePT1HCurrent capacity (use last data point)Ingress``TotalPT6HData volume inEgress``TotalPT6HData volume out

All 7 accounts can be queried in parallel (7 accounts x 9 metrics = 63 calls, within the 30-call batch cap when split into 3 batches).

Correlation analysis — when analyzing metrics together:

• High E2E latency + low server latency = network issue, not storage
• High E2E latency + high server latency = storage backend issue (throttling, overloaded partition)
• High error transactions + normal latency = client errors (auth, not-found), not performance
• High error transactions + high latency = throttling (503) or backend degradation
• Availability drop + high error transactions = sustained storage issue
• High ingress + high latency = large writes saturating bandwidth
• Dormant (all empty) = no traffic, investigate if expected or orphaned

Step 3b: Resource Logs

For each selected account, query Storage resource logs using amgmcp_query_resource_log. Keep time range to 1-2 days.

Log Query 1: Failed requests by status code

kql

StorageBlobLogs
| where TimeGenerated between (datetime(<START>) .. datetime(<END>))
| where StatusCode >= 400
| summarize count() by StatusCode, StatusText, bin(TimeGenerated, 1h)
| order by TimeGenerated asc

Log Query 2: High latency operations

kql

StorageBlobLogs
| where TimeGenerated between (datetime(<START>) .. datetime(<END>))
| where ServerLatencyMs > 100
| summarize count(), avg(ServerLatencyMs), max(ServerLatencyMs) by OperationName
| order by count_ desc
| take 20

Log Query 3: Error distribution by operation

kql

StorageBlobLogs
| where TimeGenerated between (datetime(<START>) .. datetime(<END>))
| where StatusCode >= 400
| summarize count() by OperationName, StatusCode, StatusText
| order by count_ desc
| take 30

Log Query 4: Request volume trend

kql

StorageBlobLogs
| where TimeGenerated between (datetime(<START>) .. datetime(<END>))
| summarize
    TotalRequests=count(),
    FailedRequests=countif(StatusCode >= 400),
    ThrottledRequests=countif(StatusCode == 503),
    AvgLatencyMs=round(avg(ServerLatencyMs), 2)
    by bin(TimeGenerated, 1h)
| order by TimeGenerated asc

Log Query 5: Authentication failures

kql

StorageBlobLogs
| where TimeGenerated between (datetime(<START>) .. datetime(<END>))
| where StatusCode == 403
| summarize count() by AuthenticationType, CallerIpAddress, UserAgentHeader
| order by count_ desc
| take 20

Note: If StorageBlobLogs returns no data, diagnostic settings may not be configured. Try StorageTableLogs or StorageQueueLogs. If none return data, note it and skip logs for that account.

---

Classification#

SeverityCriteriaCRITICALAvailability avg < 99.0%WARNINGAvailability avg < 99.9%, OR SuccessE2ELatency avg > 50ms, OR sustained latency > 20ms for 6+ hours, OR sustained error transactions across multiple time windowsDORMANTAll metrics return empty timeSeries (no traffic in scan period)HEALTHYAll metrics within normal ranges

---

Analysis Guidance#

For known patterns, deep-dive queries, and correlation techniques, see reference/analysis-patterns.md.

For optional deep-dive queries, see reference/deep-dive-queries.md.

---

Output Format#

Present a summary report with these sections:

1. Fleet Inventory#

Account count by region and subscription. Flag any accounts not in "Succeeded" provisioning state.

2. Pulse Check Summary#

Fleet-wide summary from the storage_summary pulse check:

• Total accounts scanned and scan status
• Key fleet-wide metrics (availability, latency, error counts, capacity)
• Breakdown by health status (healthy / warning / critical / dormant)
• Which accounts were selected for deep dive and why each was chosen

3. Deep Dive Findings (Top 7)#

For each selected account:

• Metric timeline: Availability, E2E latency, server latency, transactions over the scan window
• Anomaly characterization: spike vs sustained, onset time, duration, recovery
• Correlation analysis: which metrics moved together
• Error rate: error transactions / total transactions percentage

4. Resource Log Findings#

For each deep-dived account:

• Error summary: top error status codes, affected operations, throttling timeline
• Latency analysis: slow operations by type, server vs E2E latency comparison
• Authentication failures: if any, source IPs and user agents

5. Known Issue Cross-Reference#

Compare findings against memory/amg-check-storage-account/report.md. For each known bug, state: still active / improving / worsening / resolved.

6. Action Items#

Prioritized list:

• Critical: accounts with availability < 99.0%
• High: accounts with sustained error transactions, high latency, or throttling
• Medium: accounts with elevated latency, intermittent errors, or capacity concerns
• Low: dormant accounts (investigate if orphaned), informational items

---

Update Known Issues#

After presenting findings, update memory/amg-check-storage-account/report.md:

1. Read the current file (create if it doesn't exist).
2. Update status of existing bugs based on today's telemetry.
3. Add new bugs with: severity, account name, region, metric evidence, log evidence, root cause, recommended action.
4. Update the "Updated" date in the header.

Only add genuine issues: sustained availability drops, persistent throttling, high error rates, or latency degradation.

---

Error Handling#

See ${CLAUDE_SKILL_DIR}/reference/error-handling.md for the full recovery table.

---

Reference#

• Storage resource type: Microsoft.Storage/storageAccounts
• ARM ID template: /subscriptions/{SUB}/resourceGroups/{RG}/providers/Microsoft.Storage/storageAccounts/{name}
• Resource log tables: StorageBlobLogs (primary), StorageTableLogs, StorageQueueLogs (alternatives)
• Key status codes: 503 (throttling), 403 (auth failure), 404 (not found), 409 (conflict)
• UsedCapacity: PT1H interval only
• Known issues: memory/amg-check-storage-account/report.md
• Configuration: memory/amg-check-storage-account/config.md

---

First-Run Setup#

Run only when Config shows NOT_CONFIGURED. After completing, return to the Workflow above.

1. Discover Datasource UID: Call amgmcp_datasource_list. Filter type == "grafana-azure-monitor-datasource". Prefer uid == "azure-monitor-oob" if multiple match. Abort if zero match.

2. Discover Subscription ID(s): Run this Resource Graph query to list all subscriptions with storage accounts, then present the results as a table and ask the user which subscription(s) to use:

text

resources
| where type == 'microsoft.storage/storageaccounts'
| join kind=inner (
    resourcecontainers
    | where type == 'microsoft.resources/subscriptions'
    | project subscriptionId, subscriptionName=name
) on subscriptionId
| summarize StorageAccounts=count() by subscriptionId, subscriptionName
| order by StorageAccounts desc

Present the results as a table with columns: Subscription Name, Subscription ID, Storage Accounts. Then ask the user: "Which subscription ID(s) should I configure for this health check?"

3. Write config: Write memory/amg-check-storage-account/config.md:

markdown

# amg-check-storage-account Configuration

User-specific values for the Storage Account health check skill.
This file is auto-generated on first run and can be edited manually.

## Azure Monitor Datasource
- **UID**: {discovered_uid}
- **Name**: {discovered_name}

## Subscriptions
- {subscription_id_1}
- {subscription_id_2}

4. Confirm: Show the resolved config and ask for confirmation before proceeding.