Skip to content

EXERCISE_05_PLAN_FEATURE.md

Latest commit

 

History

History
78 lines (51 loc) · 2.54 KB

EXERCISE_05_PLAN_FEATURE.md

File metadata and controls

78 lines (51 loc) · 2.54 KB

Exercise 05 — Plan the Ticket Assignment Feature

Duration: 15 minutes | Feature: /model, /plan, Custom Agent | Goal: Invoke the custom agent, select the right AI model, and use /plan to generate a structured, phase-by-phase implementation plan for the ticket assignment feature before a single line of code is written.

Background

The Help Desk CRM has no way to assign tickets to team members. This exercise uses /plan to produce a reviewed implementation plan — covering model, schema, service, route, and tests changes — before any code is delegated. Separating planning from coding catches scope issues early.

Prerequisites

Step 1 — Start a Session and Invoke the Agent

copilot
/agent list
@helpdesk-crm-agent Hello! I need your help planning a new feature.

Step 2 — Select an AI Model

/model
/model gpt-4

Guidelines: gpt-4 / claude-3-opus for multi-file features; gpt-4-turbo / claude-3-sonnet for most tasks.

Step 3 — Generate the Implementation Plan

@helpdesk-crm-agent /plan

Plan the "Assign" feature for the Helpdesk CRM.

Requirements:
1. Add assigned_to (string) and assignee_role (enum) fields to the Ticket model
2. Roles: Developer, Test Engineer, DevOps Engineer
3. PUT endpoint /tickets/{id}/assign
4. Update Pydantic schemas with assignment fields
5. Role validation with HTTP 422 on invalid input
6. Unit tests: valid assignment, invalid role, non-existent ticket

Context:
- Architecture: Route → Service → Repository → Model
- Stack: FastAPI, SQLAlchemy, Pydantic v2, pytest, PEP 8 + type hints
- Files: app/models/, app/schemas/, app/api/, app/services/, app/repositories/

Output: numbered phases, files to change, key code changes, test cases

Review the plan — it should cover at least five phases (Model → Schema → Repository/Service → Route → Tests).

Key Takeaways

  • /plan produces a reviewable plan before code is written — use it to catch missing layers or wrong file targets.
  • Selecting a powerful model (gpt-4) for planning improves phase coverage for multi-file features.
  • Keep the plan open in the session; you will pass it directly to /delegate in Exercise 06.

Next: Exercise 06 — Delegate Implementation & Review the PR