YOU GUYS — Copilot CLI just went GA and it’s time we talk about what this actually means because the upgrade path they've built is absolutely wild! It's no longer a side project; it's an official extension installable via `--install` on Codespaces, GitHub Desktop, and even VS Code. The real shift isn't just "CLI now exists," but that the whole model has moved from one monolithic prompt to task-specific agents through a brand-new `--custom` flag. Instead of writing a 1024-character system prompt that tries to force Copilot to be your refactoring wizard, bug hunter, and test writer all at once — which is how hallucinations happen — you now define multiple mini-agents each with its own narrow mission. One agent gets the "you are an expert reviewer" instruction set; another becomes a unit-test generator that only writes Jest boilerplate; another generates commit messages based on staged changes. You can run them side by side in the same repo because their system prompts stay isolated from one another!
Think about why this matters for actual workflow instead of just being "cool new feature." When you need to refactor a module, you don't ask Copilot; you invoke your custom `refactor` agent that knows exactly which files and naming conventions are relevant. When you have a bug report come in from users, you fire up the `bug-reporter` agent preloaded with triage rules rather than asking "how do I fix this?" You can even pipe output between them — one agent drafts code and another reviews it as a second pass. This is the transition from AI as an autocomplete to AI as a multi-agent system where each task has its own bounded context, which dramatically reduces hallucination because no single prompt is asked to be everything at once. If you're still typing full English prompts into your terminal for every request, stop! Build your custom agents, give them specialized instructions per use case, and treat the CLI as a suite of dedicated tools rather than one general-purpose assistant — that's where the real productivity lives.
Source: https://github.blog/ai-and-ml/github-copilot/from-one-off-prompts-to-workflows-how-to-use-custom-agents-in-github-Copilot-CLI/
Think about why this matters for actual workflow instead of just being "cool new feature." When you need to refactor a module, you don't ask Copilot; you invoke your custom `refactor` agent that knows exactly which files and naming conventions are relevant. When you have a bug report come in from users, you fire up the `bug-reporter` agent preloaded with triage rules rather than asking "how do I fix this?" You can even pipe output between them — one agent drafts code and another reviews it as a second pass. This is the transition from AI as an autocomplete to AI as a multi-agent system where each task has its own bounded context, which dramatically reduces hallucination because no single prompt is asked to be everything at once. If you're still typing full English prompts into your terminal for every request, stop! Build your custom agents, give them specialized instructions per use case, and treat the CLI as a suite of dedicated tools rather than one general-purpose assistant — that's where the real productivity lives.
Source: https://github.blog/ai-and-ml/github-copilot/from-one-off-prompts-to-workflows-how-to-use-custom-agents-in-github-Copilot-CLI/