patterns

---
name: "patterns"
description: "Index of repeatable implementation patterns for Copilot Studio agents. When a request may need a best-practice architecture or reusable pattern for building an agent capability, retrieve this index before deciding what detailed guidance is relevant. Do not decide from this frontmatter alone; use the index summaries, then open only the specific pattern file if needed. Do not use for general knowledge sources or topic creation."
license: "MIT"
---

# Copilot Studio — Repeatable Patterns

Reference architectures for building specific capabilities in MCS agents. Each pattern is a proven, end-to-end implementation guide.

**Only read the file relevant to the current task** — do NOT read all files.

## JIT Glossary → [jit-glossary.md](jit-glossary.md)

Automatically loads a CSV of customer-specific acronyms and terminology into a global variable (`Global.Glossary`) on the first user message. The orchestrator uses it to silently expand acronyms before searching knowledge sources — improving retrieval quality without the user having to explain internal jargon.

**Read this pattern when:**
- The user wants to add a glossary, acronym list, or terminology table
- Knowledge search quality is poor because the agent doesn't understand internal abbreviations
- The user asks about loading CSV/text data from SharePoint into a variable at conversation start

## JIT User Context → [jit-user-context.md](jit-user-context.md)

Loads the current user's Microsoft 365 profile (country, department, display name, etc.) into global variables on the first user message. The orchestrator uses these to personalize answers — e.g., returning the correct country-specific WFH policy without asking the user where they are.

**Read this pattern when:**
- The user wants country-aware, department-aware, or role-aware answers
- The agent needs to call the M365 Users connector (`GetMyProfile` / `UserGet_V2`)
- The user asks about personalizing responses based on who is chatting

## Orchestrator-Generated Variables → [orchestrator-variables.md](orchestrator-variables.md)

Uses `AutomaticTaskInput` to let the orchestrator's LLM classify or extract structured data from the user's message at orchestration time — no extra AI Prompt call, no extra latency, no extra cost. The primary use case is routing knowledge searches by category or country inside an `OnKnowledgeRequested` topic.

**Read this pattern when:**
- The user needs to route knowledge searches to different sources based on query category or country
- The user wants to classify user intent without an explicit question or AI Prompt
- The user asks about `AutomaticTaskInput`, `shouldPromptUser`, or orchestrator-generated inputs

## Combining Patterns

You can combine more than one pattern. For example, when using both glossary and user context, merge them into a **single** `conversation-init` topic rather than creating separate OnActivity topics. Use the template at `${CLAUDE_SKILL_DIR}/../../templates/topics/conversation-init.topic.mcs.yml`. The individual files explain the details.