GitHub enters the terminal agent wars with Copilot CLI

yazar

GitHub has quietly been building the most compelling answer to Claude Code and OpenAI’s Codex CLI — and it just arrived as a standalone product. Copilot CLI is a terminal-native coding agent that lives outside VS Code, competes head-on with Anthropic and OpenAI’s standalone tools, and leans hard into one advantage neither rival can match: native GitHub integration.

“Less // TODO: more done. Run a GitHub-native agent in your terminal that works directly with your issues and pull requests, executes across a /fleet of parallelized subagents, and carries you from /plan to merged code.”

What actually changed

GitHub Copilot has always lived inside editors. With the Copilot CLI, it ships as a first-class standalone terminal agent — comparable in scope to Claude Code (Anthropic) and Codex CLI (OpenAI):

  • Install in one line: curl -fsSL https://gh.io/copilot-install | bash or npm install -g @github/copilot
  • Included in all plans — Free, Pro, Pro+, Business, Enterprise — no separate billing
  • /plan mode: Shift+Tab to outline work before touching code; Copilot proposes, you approve
  • /fleet: Spin up parallel subagents across multiple models simultaneously — compare results, pick the best
  • /model switching: Change foundation models mid-session (Anthropic, Google, OpenAI all supported)
  • GitHub-native MCP: Built-in access to your issues, branches, PRs and labels — no context hunting
  • /delegate: Create a branch, implement a change, open a PR — one command
  • AGENTS.md support: Define custom instructions and tool access per repo
  • Works within existing guardrails: Branch protections, required checks, org policies all apply

The workflow they’re pitching: start in the CLI with /plan, execute with /fleet, refine in /IDE (VS Code), ship via /delegate. The CLI is the on-ramp; the rest of your toolchain stays intact.

Why this matters

The standalone terminal agent market just got a lot more crowded — and GitHub’s entry has structural advantages.

Claude Code is brilliant but Anthropic-only. Codex CLI is powerful but tethered to OpenAI’s models. Copilot CLI is model-agnostic and GitHub-native, which means it can do things the others can’t: pull context from issues, enforce your branch protections automatically, and hand off directly to your existing CI/CD. You don’t swap in a new workflow — you extend the one you already have.

The /fleet feature is also worth watching. Running parallel subagents across different models on the same task and converging on a decision-ready result is genuinely new territory for a shipping product.

What to do

  • If you’re on any Copilot plan: Install it now. npm install -g @github/copilot. It’s already included.
  • Business/Enterprise teams: Your admin needs to enable Copilot CLI in org settings first.
  • Try the Skills exercise: github.com/skills/create-applications-with-the-copilot-cli — hands-on walkthrough from scratch.
  • Evaluate against Claude Code and Codex: If you’re already running one of those, Copilot CLI’s GitHub-native MCP integration is the main differentiator to test in your actual workflow.

Source: The New Stack · GitHub Copilot CLI

✏️ Drafted with KewBot (AI), edited and approved by Drew.