If Copilot ever felt "different" between VS Code and terminal sessions, you are not imagining it. The same words can mean different things depending on the surface. Today we will make four terms easy to remember: custom instructions, prompt files, custom agents, and agent skills.
Availability: All GitHub Copilot plans including Free.
The one-minute mental model
Think of Copilot customization like a team setup. Instructions are your team rules, prompt files are repeatable playbooks, custom agents are specialized teammates, and skills are what those teammates can do. Once you separate these roles, the platform differences become much easier to manage.
1) Custom instructions: your shared rules
Custom instructions tell Copilot how to behave in your project. They are best for stable guidance: coding style, architecture boundaries, content standards, and preferred workflows.
Simple example: in .github/copilot-instructions.md, define "Use semantic HTML, keep dates in DD Month YYYY,
and always include an availability blockquote in posts."
Use case: every collaborator gets consistent Copilot output without rewriting the same expectations in each chat.
2) Prompt files: reusable workflows in IDE chat
Prompt files are reusable commands that package a full workflow. In IDE chat, you invoke them with
/prompt-name and run a guided process, like audit then implement.
Simple example: .github/prompts/blog-audit.prompt.md checks one post for required sections,
broken structure, and docs links, then asks "Shall I implement fixes?".
Use case: repeat quality checks across many posts with the same process and output format.
3) Custom agents: role-focused helpers
Custom agents are specialized Copilot personas tuned for a type of work. Instead of one general assistant doing everything, you can route tasks to the best specialist.
Simple example: use a writing-focused agent to draft blog structure and a code-review-focused agent to verify HTML consistency.
Use case: split complex work so each phase uses an agent optimized for that job.
4) Agent skills: concrete capabilities
Agent skills are the actions an agent can take, such as searching files, reading content, editing files, or fetching documentation. Skills define execution power.
Simple example: an audit prompt uses search/read to find issues, then edit to apply approved fixes.
Use case: design workflows where Copilot does real multi-step work, not just one-shot answers.
VS Code vs CLI: what changes
| Customization type | VS Code / IDE | Copilot CLI |
|---|---|---|
| Project instructions | Auto-loaded from .github/copilot-instructions.md |
Auto-loaded from .github/copilot-instructions.md |
| Prompt files | Supported via /prompt-name |
Not directly invoked with slash prompts |
| Path-specific instruction files | Can be referenced in project setup | Useful for contextual CLI guidance |
| Best outcome | Interactive guided workflows | Conversational terminal execution with shared rules |
A practical setup you can copy
- Keep core conventions in
.github/copilot-instructions.mdas your source of truth. - Create IDE prompt files for repeatable workflows like audits and new-post generation.
- Add focused instruction files for CLI-specific guidance and standards.
This gives you one consistent "brain" for conventions, plus platform-specific execution layers where needed.