ICM DRI Agent

Automated triage/investigation agent for RAI DRI workflows.

It investigates ICM incidents by combining:

• ICM incident metadata/context via MCP/API
• TSG lookup from ADO wiki (Erie)
• Kusto diagnostics for correlation IDs
• LLM synthesis into root-cause summaries

What it does

For each incident, the agent can:

1. Fetch incident details, AI summary, customer impact, and related context.
2. Resolve and fetch TSG content (eng.ms URL, ADO wiki URL, or wiki path).
3. Run diagnostic Kusto queries (template-driven or custom).
4. Generate a concise root-cause summary + recommended actions.

Outputs are saved under runs/:

• Per-incident summaries: <incident_id>_summary.md
• Batch summary: <timestamp>_batch_summary.md
• Batch logs: batch.log

Requirements

• Python 3.11+
• Azure login (az login) with access to:
  • Azure OpenAI resource
  • ICM MCP/API scopes
  • Azure DevOps wiki
  • Kusto clusters/databases

Setup (Windows + Ubuntu/macOS)

1) Install dependencies

PowerShell:

python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt

Bash:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2) Authenticate

az login

3) Configure runtime

Choose one:

• .env (preferred for CI/automation)
• model.config (local dev convenience)

Minimum values:

• AZURE_OPENAI_ENDPOINT (or target in model.config)
• AZURE_OPENAI_API_KEY (or key in model.config)
• AZURE_OPENAI_MODEL / model_name

model.config.example is included as a starter.

Run

Single incident

python -m src.main 757179687

Batch mode (default)

python -m src.main --team "RAI - AACS DRI" --min-status active --min-severity 25 --lookback 7

Interactive mode

python -m src.main --interactive

Query templates

Queries.md now uses parameterized templates with placeholders:

• {{CORRELATION_ID}}
• {{START_TIME_UTC}}
• {{END_TIME_UTC}}

The agent can call:

• get_query_templates to inspect template metadata
• run_kusto_template to execute by template_name + correlation_id (+ optional time range)
• run_kusto_query for custom KQL

Notes on MCP workspace config

This runtime does not require VS Code Copilot MCP integration to execute tools. Tool calls are implemented inside the Python runtime (src/tools/*) and invoked by the agent loop.

• A VS Code MCP workspace config is not required for this runtime.
• Tool calls run in the Python agent process (src/tools/*) and are independent of editor MCP wiring.