Prompts Overview
Copilot Collections includes 13 public prompts — slash commands that trigger specific workflow actions across the full product lifecycle. Prompts are stored in .github/prompts/ and become available as /command shortcuts in VS Code chat.
How Prompts Work
Each prompt file defines:
- Agent binding — Which agent executes the command, via the
agentfield. Model selection is inferred from that agent's ownmodelarray; public prompts do not declare a separate prompt-levelmodelfield. - Description — Shown in VS Code's command palette.
- Instructions — Detailed workflow steps, required skills, and output format.
When you type /tsh-explore-materials, /tsh-implement, /tsh-review, etc. in the VS Code chat, the corresponding prompt file is loaded and executed by the bound agent.
Public Prompts
These are the user-facing commands available in VS Code chat.
📋 Product Ideation Commands
| Command | Agent | Description |
|---|---|---|
| /tsh-explore-materials | Business Analyst | Explore workshop materials before backlog extraction |
| /tsh-analyze-materials | Business Analyst | Process workshop materials into Jira-ready epics and stories |
🛠 Development Commands
| Command | Agent | Description |
|---|---|---|
| /tsh-implement | Engineering Manager | Orchestrate the full cycle: research/context gathering if needed → plan → implementation |
✅ Quality Commands
| Command | Agent | Description |
|---|---|---|
| /tsh-review | Code Reviewer | Review implementation against plan and standards |
| /tsh-review-ui | UI Reviewer | Single-pass Figma vs implementation comparison |
| /tsh-review-codebase | Architect | Comprehensive code quality analysis |
⚙️ Copilot Customization Commands
| Command | Agent | Description |
|---|---|---|
| /tsh-create-custom-agent | Copilot Orchestrator | Create a new custom agent |
| /tsh-create-custom-skill | Copilot Orchestrator | Create a new custom skill |
| /tsh-create-custom-prompt | Copilot Orchestrator | Create a new custom prompt |
| /tsh-create-custom-instructions | Copilot Orchestrator | Create custom instruction files |
🏗 Infrastructure & Cost Analysis Commands
| Command | Agent | Description |
|---|---|---|
| /tsh-audit-infrastructure | DevOps Engineer | Audit infrastructure for security gaps, cost waste, and best practices |
| /tsh-analyze-aws-costs | DevOps Engineer | AWS cost optimization and tagging compliance audit |
| /tsh-analyze-gcp-costs | DevOps Engineer | GCP cost optimization and labeling compliance audit |
Delegation via /tsh-implement
When you run /tsh-implement, the Engineering Manager automatically handles the full development cycle. Its four primary inputs are a task description, Jira ID, standalone *.research.md, or *.plan.md. Missing research or plan companions trigger preparation, never no-plan implementation. Before the first file-changing delegation in either flow, the manager requires Human approval of the exact current plan revision; automated Reviewer approval is readiness evidence only, not permission to implement. You don’t need to invoke individual agents — the orchestration is handled for you.
| Phase | Delegated To |
|---|---|
| Research (context gathering) | Context Engineer (via internal research prompt) |
| Planning (architecture) | Architect (via internal plan prompt) |
| Plan validation | Architect Reviewer stress-test (via internal review-plan prompt) |
| Backend / general code (actionable, low-risk seam) | Plan Implementor — DEFAULT |
| Backend / general code (complex, non-UI) | Software Engineer — EXCEPTION |
| Frontend with Figma | UI Engineer (via internal UI prompt) |
| E2E tests | E2E Engineer |
| LLM application prompts | Prompt Engineer (via internal engineer-prompt) |
| Kubernetes, Terraform, CI/CD, observability | DevOps Engineer |
| Repository documentation | Technical Writer |
| UI verification | UI Reviewer |
Input Format
Most prompts accept either:
- A Jira ticket ID:
/tsh-implement PROJ-123 - A task description:
/tsh-implement Add pagination to the user list API
/tsh-implement additionally accepts a standalone *.research.md file or a *.plan.md implementation plan as a primary input. A missing research or plan companion routes to preparation; it never authorizes no-plan implementation.
The agent adapts its behavior based on the input type — pulling context from Jira/Confluence for ticket IDs, or working from the description and codebase for free-form text.