🛡️ Heimdall - SRE Agent

AI-powered Site Reliability Engineering agent for incident management, root cause analysis, and reliability engineering.

Features

| Command | Description | |---------|-------------| | [TI] Triage Issue | Analyze JIRA issues, identify root causes, assign priority | | [CP] Create Postmortem | Generate RCA and COE documents using templates | | [DP] Debug Prometheus | Validate PromQL syntax, suggest optimizations | | [PA] Performance Analysis | Analyze metrics, DB query optimization, capacity planning | | [RC] Runbook Creation | Create structured SRE runbooks with validation | | [CI] CI/CD Reliability | Analyze pipeline health and recommend improvements | | [ST] SLO/SLI Tracking | Define service level objectives and monitoring | | [OH] On-Call Handoff | Generate structured handoff documentation | | [CA] Cost Analysis | Analyze infrastructure costs and optimization |

Installation

Via NPX (Recommended)

npx @chocoobar/heimdall

This installs Heimdall into your project's .agent/heimdall/ directory.

Manual Installation

1. Clone this repository
2. Copy contents to .agent/heimdall/ in your project

Activation

After installation, tell your AI assistant:

Load and activate the Heimdall agent from .agent/heimdall/agents/heimdall.md

Configuration

Edit .agent/heimdall/config.yaml:

sre:
  user_name: "your-name"
  output_folder: "_bmad-output"
  project_name: "your-project"
  language: "English"
  
  # Output directory for SRE artifacts
  sre_artifacts: "_sre-artifacts"
  
  # MCP Configuration
  mcp_jira_enabled: true
  mcp_confluence_enabled: false
  
  # Prometheus endpoint (optional)
  prometheus_endpoint: "http://prometheus:9090"

MCP Integration

Heimdall integrates with Atlassian MCP for JIRA/Confluence access.

Setup

1. Install Atlassian MCP server: atlassian-mcp-server
2. Configure in your IDE/editor settings
3. Set mcp_jira_enabled: true in config.yaml

Features with MCP

• Automatic ticket retrieval from JIRA
• Attachment reading (logs, configs, stack traces)
• Publish postmortems directly to Confluence

Fallback Mode

If MCP is unavailable, Heimdall prompts for manual input with clear guidance.

Workflows

Triage Issue

[TI] → Acquire ticket → Analyze → Root cause → Generate report

• Supports JIRA MCP or manual input
• Reads ticket attachments for logs/stack traces
• Searches codebase for error patterns (when installed in a project)

Create Postmortem

[CP] → Gather data → Generate RCA → Generate COE → Publish

• Uses 5 Whys methodology
• Generates both RCA and COE documents
• Optional Confluence publishing

Debug Prometheus

[DP] → Parse query → Analyze → Optimize → Output

• Validates PromQL syntax
• Suggests performance optimizations
• Tests against live Prometheus (if configured)

Output

All outputs are saved to {sre_artifacts}/ directory:

• postmortems/ - RCA and COE documents
• runbooks/ - Generated runbooks
• reports/ - Triage and analysis reports