azure-sre-agent

Azure SRE Agent and srectl CLI

Azure SRE Agent is an AI-powered reliability platform that automates incident response, monitoring, and operational tasks. The srectl CLI is the command-line interface for managing SRE Agent instances—creating agents, skills, tools, and scheduled tasks via YAML configurations.

Installation

# Install the Azure Artifacts credential provider (one-time)
curl -L https://raw.githubusercontent.com/microsoft/artifacts-credprovider/master/helpers/installcredprovider.sh | bash

# Install srectl (.NET 9.0 SDK required)
dotnet tool install sreagent.cli --global \
  --add-source https://pkgs.dev.azure.com/msazure/One/_packaging/SREAgentCli/nuget/v3/index.json \
  --interactive

# Verify
srectl --version

Prerequisites: .NET 9.0 SDK, Azure CLI (az login for authentication).

Key concepts

Architecture: trigger → agent → tools → connectors

The runtime flow is: a trigger (ICM incident, scheduled task, or manual chat) invokes an agent, which uses tools (KustoTool, LinkTool, or system tools) that connect to connectors (pre-authenticated endpoints for Kusto, ICM, Outlook, Teams, MCP).

Deployment order

Always deploy dependencies first: tools → agents → scheduled tasks. Agents reference tools by name, so tools must exist before agents. Scheduled tasks reference agents, so agents must exist before tasks.

YAML schema (V2)

All YAML uses api_version: azuresre.ai/v2. Two kind values: ExtendedAgent for agents, ExtendedAgentTool for tools. Use ##param## format for parameter placeholders in queries and templates—never {{param}}.

Workspace structure

your-project/
├── agents/
│   └── MyAgent/MyAgent.yaml
├── tools/
│   └── MyTool/MyTool.yaml
├── skills/
│   └── my-skill/SKILL.md
└── scheduledtasks/

Profiles

Profiles store connection settings for different SRE Agent instances. Use srectl profile create to add instances and srectl profile set to switch between them.

Command reference

Initialization and status

Command	Purpose
srectl init --resource-url <url>	Connect to an SRE Agent instance
srectl status	Show workspace health
srectl sync	Pull remote configs to local workspace
srectl profile list	List connection profiles
srectl profile create --name <n> --url <url>	Create a new profile
srectl profile set --name <n>	Switch active profile

Agent lifecycle

Command	Purpose
srectl agent create --name <n> --instructions "..."	Create agent YAML
srectl agent create --name <n> --smart --instructions "..."	Create with AI assistance
srectl agent validate --name <n>	Validate agent YAML locally
srectl agent validate --all --check-tools	Validate all agents with tool checks
srectl agent apply --name <n>	Deploy agent to server
srectl agent apply --name <n> --dry-run	Preview deployment
srectl agent test --name <n> --message "..."	Test agent interactively
srectl agent diff --name <n>	Compare local vs remote
srectl agent list	List remote agents
srectl agent delete --name <n>	Delete remote agent

Skill lifecycle

Command	Purpose
srectl skill create --name <n>	Create skill directory with templates
srectl skill apply --name <n>	Upload skill to server
srectl skill apply --name <n> --dry-run	Preview upload
srectl skill list	List remote skills
srectl skill sync --name <n>	Download skill from server
srectl skill sync --all	Download all skills
srectl skill convert --agent-name <n>	Convert agent to skill
srectl skill delete --name <n>	Delete remote skill

Tool lifecycle

Command	Purpose
srectl tool create	Create tool YAML
srectl tool validate --name <n>	Validate tool YAML
srectl tool apply --name <n>	Deploy tool to server
srectl tool show-types	List available tool types (KustoTool, LinkTool)
srectl tool show-connectors	List configured connectors
srectl tool list	List remote tools
srectl tool delete --name <n>	Delete remote tool

Apply YAML (kubectl-style)

Command	Purpose
srectl apply-yaml --file <path>	Apply single YAML file
srectl apply-yaml --file manifests/all.yaml	Apply multi-document YAML

Knowledge documents

Command	Purpose
srectl doc upload --file <path>	Upload a document
srectl doc upload --file ./docs	Upload entire folder
srectl doc search --query "..."	Search knowledge base
srectl doc get	List uploaded documents
srectl doc delete --name <n>	Delete a document
srectl doc reindex	Reindex all documents

Scheduled tasks

Command	Purpose
srectl scheduledtask create --name "..." --cron "..." --prompt "..." --agent "..."	Create scheduled task
srectl scheduledtask list	List tasks
srectl scheduledtask pause --id <id>	Pause task
srectl scheduledtask resume --id <id>	Resume task
srectl scheduledtask delete --id <id>	Delete task

Hooks (safety and governance)

Command	Purpose
srectl hook list	List hooks
srectl hook apply --file <path>	Create/update hook
srectl hook activate --thread-id <id> --name <n>	Activate for thread
srectl hook deactivate --thread-id <id> --name <n>	Deactivate for thread

MCP server

Command	Purpose
srectl mcp start	Start MCP server (stdio transport)
srectl mcp info	Show platform documentation
srectl mcp info --topic <topic>	Show specific topic
srectl mcp info --all	Show all documentation

Interactive and chat

Command	Purpose
srectl chat	Start interactive chat with SRE Agent
srectl interactive	Start guided mode

Agent YAML schema

api_version: azuresre.ai/v2
kind: ExtendedAgent
metadata:
  name: my-agent                      # kebab-case
spec:
  instructions: |-                    # Multi-line agent instructions
    You are a specialist that...
  handoffDescription: 'Short description for routing'
  handoffs: []                        # Other agents this can delegate to
  tools:
    - tool-name-1
    - tool-name-2
  maxReflectionCount: 0
  customReflectionNote: ''
  commonPrompts: []
  enableVanillaMode: false

KustoTool YAML schema

Use ##param## for placeholders. Use type: string and target: dictionary:args:string for all parameters.

api_version: azuresre.ai/v2
kind: ExtendedAgentTool
metadata:
  name: get-service-errors
spec:
  type: KustoTool
  connector: my-kusto-connector       # From srectl tool show-connectors
  toolMode: Auto
  description: |-
    Purpose: Get error counts for a service
    Usage: Call with subscriptionId and hours
    Output Format: ErrorCount by hourly bins
  database: TelemetryDB
  query: |-
    let _sub = '##subscriptionId##';
    let _hours = toint('##hours##');
    ServiceLogs
    | where SubscriptionId == _sub
    | where Timestamp > ago(_hours * 1h)
    | where Level == 'Error'
    | summarize ErrorCount=count() by bin(Timestamp, 1h)
  parameters:
    - name: subscriptionId
      type: string
      description: Azure subscription ID
      required: true
      target: dictionary:args:string
    - name: hours
      type: string
      description: Hours to look back
      required: true
      target: dictionary:args:string

LinkTool YAML schema

api_version: azuresre.ai/v2
kind: ExtendedAgentTool
metadata:
  name: open-dashboard
spec:
  type: LinkTool
  toolMode: Auto
  description: Open the Grafana dashboard for a resource
  template: https://grafana.example.com/d/##dashboardId##?var-resource=##resourceName##
  parameters:
    - name: dashboardId
      type: string
      description: Grafana dashboard ID
      required: true
      target: dictionary:args:string
    - name: resourceName
      type: string
      description: Resource name to filter
      required: true
      target: dictionary:args:string

Deployment workflow

1. Check connectors: srectl tool show-connectors
2. Create tools: srectl apply-yaml --file tools/my-tool.yaml
3. Create agent: srectl apply-yaml --file agents/my-agent.yaml
4. Test: srectl agent test --name my-agent --message "check system health"
5. Schedule (if needed): srectl scheduledtask create --name "daily-check" --cron "0 9 * * *" --agent "my-agent" --prompt "Run daily health check"

MCP server configuration

Copilot CLI

Add to ~/.copilot/mcp-config.json:

{
  "mcpServers": {
    "srectl": {
      "type": "stdio",
      "command": "/Users/brett/.dotnet/tools/srectl",
      "args": ["mcp", "start"],
      "env": {
        "DOTNET_ROOT": "/Users/brett/.dotnet"
      }
    }
  }
}

VS Code

Add to .vscode/settings.json:

{
  "mcp.servers": {
    "srectl": {
      "command": "srectl",
      "args": ["mcp", "start"]
    }
  }
}

Common patterns

Deploy a capacity management skill

srectl init --resource-url https://my-agent.1abc.eastus2.azuresre.ai
srectl skill apply --name azure-capacity-management
srectl apply-yaml --file sre-agent/subagent/capacity-manager.yaml
srectl doc upload --file docs/operations/

Create a scheduled quota check

srectl scheduledtask create \
  --name "daily-quota-check" \
  --cron "0 9 * * *" \
  --agent "capacity-manager" \
  --prompt "Check quota utilization across all production subscriptions and alert if any VM family exceeds 80%"

Validate before deploying

srectl agent validate --all --check-tools
srectl tool validate --all

Learn more

Use srectl mcp info --topic <topic> for detailed documentation on any topic:

Topic	Description
overview	Platform overview
architecture	Trigger → agent → tool flow
cli	CLI command structure
agents	Agent configuration
tools	Tool configuration
triggers	Trigger types
subagents	Subagent and handoff concepts
scheduled-tasks	Scheduled task configuration
quickstart	Getting started

Source: Azure SRE Agent documentation