Some copilot CLI tips

Been trying this for a few days, here are some tips and ideas:

• Kind of nice that this now aligns the names for both the @copilot agent in Github.com and the CLI agent.
• Launch only from trusted directories only; approvals are required unless you explicitly allow tools. Consider containers/VMs for extra isolation.
• Default model is Claude Sonnet 4. Switch using env var: COPILOT_MODEL=gpt-5 copilot
• or per run COPILOT_MODEL=gpt-5 copilot etc.
• Use it one off: copilot -p "List my open PRs"
• pass --allow-all-tools for non-interactive runs
• copilot -p "Create a GitHub Actions workflow that runs ESLint on PRs and fails the check on any errors."
• Use it with MCPs
  • copilot -p "Use the GitHub MCP server to find good first issues for a new teammate in OWNER/REPO"
• Next I am working on pairing the CLI with profiling tools in the Chrome DevTools MCP to run entire performance testing suites.