Agent 365 Synthetic Worker: Sales Ops Bot

This repository contains the source code for a Sales Operations Digital Employee, built using the Microsoft Agent 365 SDK and Model Context Protocol (MCP).

Current Status (2026-02-16)

• ✅ Deployment healthy in Azure Container Apps (<container-app-revision> active, 100% traffic)
• ✅ Health endpoint reports status: ok, agent_initialized: true, MCP state ready
• ✅ Tests passing locally (pytest -q)
• ✅ Report/email UX updates live (summary normalization, readable header, sentiment + expansion sections)

\uD83D\uDCC2 Documentation

Document	Purpose
SETUP.md	Start Here. Prerequisites and "Definition of Ready".
PLAN.md	Step-by-step technical execution plan.
ROADMAP.md	Strategic milestones from "Intern" to "Enterprise".
SPEC-SyntheticWorker.md	Detailed architecture and functional specification.
HANDOVER.md	Current state of the codebase vs. the desired state.
docs/a365-guided-setup.md	AI-guided a365 CLI setup reference (from Agent365-devTools).
docs/TROUBLESHOOTING-A365.md	Known issues and fixes (ISS-001 – ISS-029).
docs/INCIDENT-2026-02-MEETING-REPORT.md	Post-incident write-up for transcript/report stabilization and runbook learnings.
PUBLIC-ANONYMIZATION.md	Public-safe placeholder rules and publish workflow.

\uD83D\uDE80 Quick Start

1. Prerequisites: Ensure you have Python 3.11+, .NET 8 SDK, and Agent 365 CLI installed (dotnet tool install --global Microsoft.Agents.A365.DevTools.Cli --prerelease). See SETUP.md.
2. Install: pip install -e .
3. Configure: Copy .env.template to .env and add your LLM keys.
4. Lifecycle flow: Follow Agent 365 development lifecycle steps: build/run → config → blueprint → deploy/publish.

🤖 AI-Guided Setup (Recommended)

The fastest way to set up or redeploy this agent is using the AI-guided setup prompt from the official Agent365-devTools repository. This is an AI-agent-compatible runbook that walks a coding agent (e.g. GitHub Copilot, Cursor, Claude) through the full a365 CLI pipeline end-to-end.

How to Use

1. Copy the a365-setup-instructions.md into your AI agent's context (paste it as a prompt, or save it locally and reference it).
2. The prompt instructs the agent to execute 5 steps autonomously:

Step	What It Does
1. Verify CLI	Installs/updates the a365 CLI (.NET global tool)
2. Prerequisites	Validates Azure CLI login, Entra ID roles, custom client app registration with 5 required Graph permissions, and language-specific build tools
3. Configure	Gathers inputs (resource group, location, agent name, manager email), derives naming conventions, creates a365.config.json, imports via a365 config init
4. Setup	Runs a365 setup all — provisions infrastructure, creates Entra ID blueprint, configures permissions, registers messaging endpoint
5. Publish & Deploy	Reviews manifest, runs a365 publish + a365 deploy, configures agent in Teams Developer Portal, creates agent instance

Why This Matters

• Repeatable: Same prompt produces a consistent setup every time, across any Agent 365 project
• Self-healing: The prompt includes validation gates at every step, error handling, and troubleshooting for common failures (RBAC, quota limits, Graph permission errors, dev tunnel issues)
• Project-agnostic: Detects .NET, Node.js, or Python projects automatically and validates the correct build toolchain
• Split-tenant aware: Handles scenarios where Azure infrastructure and M365 identity live in different tenants (as in this project)
• Idempotent: Steps can be safely re-run after fixing issues — the CLI skips existing resources

Key Prerequisites (from the guided setup)

The custom client app registration in Entra ID requires these 5 delegated permissions with admin consent:

Permission	Purpose
AgentIdentityBlueprint.ReadWrite.All	Manage Agent 365 Blueprints
AgentIdentityBlueprint.UpdateAuthProperties.All	Update Blueprint auth properties
Application.ReadWrite.All	Create and manage Azure AD applications
DelegatedPermissionGrant.ReadWrite.All	Grant delegated permissions
Directory.Read.All	Read directory data

Self-Hosted vs Azure-Hosted

This project uses a self-hosted deployment model (needDeployment: false) since we deploy to Azure Container Apps rather than the CLI's built-in Azure App Service. When running the guided setup:

• Choose "No" when asked about creating a web app in Azure
• Set the messagingEndpoint to your Container App URL: https://<your-app>.azurecontainerapps.io/api/messages
• The CLI still provisions the Entra ID blueprint, permissions, and agent identity

Local Reference

A local reference is maintained at docs/a365-guided-setup.md.

Source: microsoft/Agent365-devTools — keep in sync as the CLI evolves.

Architecture

The agent runs as a headless service:

1. Triggers on MeetingEnded.
2. Fetches transcripts via Microsoft Graph client (GraphTranscriptClient) using event → online meeting resolution and transcript content retrieval.
3. Computes structured insights using LLM (summary, executive summary, tasks, sentiment, churn risk, customer satisfaction, CRM update hints).
4. Generates a post-meeting Word report autonomously via Word MCP when the tool is active.
5. Pushes account-note updates autonomously via Dynamics/Dataverse MCP when available, with a safe local Dynamics stub fallback.
6. Delivers summaries via Graph and/or MCP tools depending on enabled server set (ToolingManifest.json currently enables Word, SharePoint Lists, Knowledge, Planner), with local tool fallback paths for resilience.

Deployment

Docker (Local Build)

docker build -t sales-ops-bot .
docker run -p 8000:8000 --env-file .env sales-ops-bot

Azure Container Apps

1. Build & Push:

   az acr build --registry <your-registry> --image sales-ops-bot:latest .

2. Deploy:

   az containerapp create --yaml deployment.yaml --resource-group <your-rg>

Project Structure

• src/: Core agent logic.
• tests/: Unit and integration tests.
• docs/: Reference documentation (MCP Registry).
• Dockerfile: Container definition.
• deployment.yaml: Azure manifest.

Repository Hygiene

• Canonical lifecycle/runbooks: docs/a365-guided-setup.md and docs/TROUBLESHOOTING-A365.md.
• Generated diagnostics/artifacts (logs, build outputs, captured doc extracts) are excluded via .gitignore and should not be treated as source of truth for implementation decisions.
• If a generated artifact conflicts with canonical docs, update the canonical docs and regenerate artifacts as needed.