Skip to content

customize-copilot-prompt-guide.md

Latest commit

 

History

History
303 lines (196 loc) · 12.2 KB

customize-copilot-prompt-guide.md

File metadata and controls

303 lines (196 loc) · 12.2 KB

Customize Copilot with /kit-customize-copilot

A user-facing guide to the preferred public entrypoint for Copilot customization in this repository

Table of Contents

What /kit-customize-copilot is for

/kit-customize-copilot is the preferred public entrypoint for Copilot customization work in this repository.

Use it when you want Copilot to help you:

  • create or improve project-wide Copilot guidance,
  • add instructions for a specific language, framework, or file type,
  • create a custom agent,
  • create a reusable prompt or skill workflow,
  • customize commit message behavior,
  • or audit your current Copilot setup and recommend gaps.

In plain English: if your goal is "make Copilot work better for this project or workflow", this is usually the best place to start.

You do not need to decide up front whether the answer should be an instruction file, a prompt, a skill, an agent, or a combination of artifacts. Describe the outcome you want, and let the orchestration layer figure out the likely implementation path.

How it routes to kit-copilot-engineering-team

When you run /kit-customize-copilot, the prompt routes your request to kit-copilot-engineering-team.

That team acts as the repository's Copilot customization orchestrator. Its job is to:

  • understand the outcome you want,
  • inspect the relevant repository context,
  • classify the request,
  • and choose the most appropriate internal workflow for the job.

That means you do not need to pick an internal kit-copilot-create-* asset yourself.

Instead of asking yourself questions like:

  • "Do I need a prompt or a skill?"
  • "Should this be project-wide instructions or tech-specific instructions?"
  • "Do I need an agent, a prompt, or both?"

...you can simply describe the result you want.

Direct agent usage is still possible if you prefer selecting kit-copilot-engineering-team from the agent picker, but it is not the primary documented path here. The recommended public starting point is still /kit-customize-copilot.

How to use it in Copilot Chat

  1. Open Copilot Chat in VS Code.
  2. Type /kit-customize-copilot.
  3. Add a plain-language request describing the customization outcome you want.
  4. Send the message and continue the conversation if Copilot asks for small clarifications.

Basic shape:

/kit-customize-copilot <describe the Copilot customization outcome you want>

Example:

/kit-customize-copilot Create project-wide Copilot guidance for this repository so suggestions follow our TypeScript, testing, and architecture conventions.

You can keep it short, but better requests usually include:

  • the outcome you want,
  • the scope (whole repo, React files, commit messages, custom agent, and so on),
  • relevant technologies,
  • and any important constraints or preferences.

How to describe the outcome you want

The easiest way to get a good result is to describe what you want Copilot to do better, not which internal skill you think should be used.

Good request ingredients:

  • Desired outcome — What should improve?
  • Scope — Whole repository, one tech stack, one file family, one workflow, one persona.
  • Context — Languages, frameworks, architecture, team conventions.
  • Constraints — Required style, naming, review depth, output format, or guardrails.

A simple pattern that works well:

/kit-customize-copilot I want Copilot to <desired behavior/outcome> for <scope>. This project uses <tech/context>. Please create or update the customization artifacts that fit best.

Good examples of outcome-based requests

  • "I want Copilot to follow our repo-wide conventions when editing code."
  • "I want better instructions for React components and test files."
  • "I need a custom agent for backend API design reviews."
  • "I want a reusable workflow for generating migration plans."
  • "I want commit messages to follow our release process."
  • "Please audit the current Copilot setup and tell me what's missing."

What you usually do not need to do

You typically do not need to say things like:

  • "Use the create-agent skill"
  • "Invoke the create-instruction skill"
  • "Route this through a specific internal worker"

If you already know the artifact type, you can mention it. But if you do not, plain-language intent is enough. The orchestrator is expected to classify the request and choose the likely internal path.

Copy-paste-ready examples by user intent

The examples below are written to be pasted directly into Copilot Chat. The listed results are likely or typical outcomes, not hard guarantees about the exact internal routing.

Project-wide Copilot guidance

Use this kind of request when you want broad, repository-level Copilot behavior.

Example requests:

/kit-customize-copilot Create project-wide Copilot guidance for this repository so suggestions follow our TypeScript, testing, and architectural conventions.

/kit-customize-copilot Review this repo and add or improve the top-level Copilot instructions so the AI stays aligned with our stack, folder structure, and team rules.

Likely outcome:

A typical outcome is that the orchestrator reviews the repository and creates or updates broad Copilot guidance that applies across the project.

Likely artifacts:

  • .github/copilot-instructions.md
  • possibly supporting .github/instructions/*.instructions.md files if the repo needs more targeted guidance

File-type or tech-specific instructions

Use this when the customization should apply only to a certain language, framework, layer, or file pattern.

Example requests:

/kit-customize-copilot Add Copilot instructions for React and TypeScript component files so generated code follows our props, styling, and testing patterns.

/kit-customize-copilot Create targeted guidance for API route files and database migration files. I want Copilot to be stricter about validation, error handling, and rollback safety.

Likely outcome:

The orchestrator will typically classify this as a targeted-instructions request and produce technology-specific or file-pattern-specific guidance rather than only broad repo-wide guidance.

Likely artifacts:

  • one or more .github/instructions/*.instructions.md files
  • sometimes a refinement to .github/copilot-instructions.md if shared context also needs tightening

Custom agent creation

Use this when you want a specialized Copilot persona with a distinct role, workflow, or tool posture.

Example requests:

/kit-customize-copilot Create a custom agent for security-focused code reviews in Node.js services. I want it to be strict about auth, secrets, validation, and dependency risk.

/kit-customize-copilot I need a backend architecture agent that helps design APIs, data models, and service boundaries for Go and TypeScript projects.

Likely outcome:

A common outcome is that the orchestrator defines a new specialized agent with a clear persona, workflow, and usage guidance.

Likely artifacts:

  • .github/agents/<agent-name>.agent.md
  • sometimes related instructions or documentation if the agent benefits from shared project context

Reusable prompt or skill workflow creation

Use this when you want a repeatable workflow for a recurring task.

Example requests:

/kit-customize-copilot Create a reusable Copilot prompt for generating API implementation plans from selected route definitions and project conventions.

/kit-customize-copilot Build a reusable workflow for scaffolding database migration review checklists. Choose a prompt or skill format based on what fits best.

/kit-customize-copilot I need a repeatable workflow for turning a selected specification into implementation tasks. Please create the best customization mechanism for that.

Likely outcome:

The orchestrator will often choose a reusable task workflow artifact. Depending on the complexity, that may be a prompt, a skill, or a combination of lightweight supporting assets.

Likely artifacts:

  • .github/prompts/<name>.prompt.md
  • or .github/skills/<skill-name>/SKILL.md
  • possibly supporting resource files if the workflow is richer than a single prompt

Commit message customization

Use this when you want Copilot-generated commit messages to follow a team or repository standard.

Example requests:

/kit-customize-copilot Set up commit message instructions so Copilot writes Conventional Commits with a scope based on the changed area of the repository.

/kit-customize-copilot Customize our commit message guidance so releases, fixes, and refactors follow a consistent format with short summaries and useful bodies.

Likely outcome:

A typical outcome is commit-message guidance tailored to your team's preferred structure and terminology.

Likely artifacts:

  • .github/commit-message-instructions.md

Workspace audit or gap analysis

Use this when you are not sure what should be created yet and want the repository assessed first.

Example requests:

/kit-customize-copilot Audit this workspace and tell me which Copilot customization files are missing, weak, duplicated, or outdated.

/kit-customize-copilot Review the current Copilot setup in this repository and recommend the highest-value improvements before creating anything.

/kit-customize-copilot Do a gap analysis of our Copilot customizations for this codebase, especially around instructions, prompts, agents, and commit message guidance.

Likely outcome:

The orchestrator will typically start with discovery and classification, then produce recommendations and may suggest or create a prioritized set of follow-up artifacts.

Likely artifacts:

  • an audit summary in chat
  • possibly recommended updates to existing .github/ customization files
  • sometimes new artifacts if the request explicitly asks the agent to implement the recommended changes

Internal building blocks vs public starting point

You may still see repository-owned kit-copilot-create-* assets in this repository. That is expected.

Those assets still matter, but they are now best understood as internal building blocks rather than the preferred public starting point.

In practice, that means:

  • they still exist as reusable procedural assets,
  • they still help power customization workflows behind the scenes,
  • but users are generally not expected to choose among them manually before starting.

The recommended public UX is:

  1. start with /kit-customize-copilot,
  2. describe the outcome you want,
  3. let kit-copilot-engineering-team classify the request and choose the likely implementation path.

If you are an advanced user and want to work directly with an agent or a lower-level artifact path, that can still be possible in some situations. It is just not the primary path this repository documents and recommends.

Quick recap

  • /kit-customize-copilot is the preferred public starting point for Copilot customization in this repository.
  • It routes to kit-copilot-engineering-team.
  • You do not need to choose an internal skill first.
  • Describe the outcome you want in plain language.
  • The orchestrator will typically decide the likely artifact or workflow path.
  • The kit-copilot-create-* assets remain useful internal building blocks, but they are not the preferred public entry surface.