COPILOT CLI ADDS A "CRITIC" AGENT AS AGENT SKILLS LAND IN VS CODE
GitHub is pushing Copilot from autocomplete to agent workflows across IDEs and the CLI, with a new Critic reviewer. The Copilot feature matrix shows Agent mode...
GitHub is pushing Copilot from autocomplete to agent workflows across IDEs and the CLI, with a new Critic reviewer.
The Copilot feature matrix shows Agent mode across major IDEs, with agent skills now in VS Code 1.108 and custom agents graduating from preview there. MCP is broadly supported, while BYOK remains in preview across IDEs. See the evolving table in GitHub’s feature matrix.
On the command line, Copilot CLI v1.0.18 adds a Critic agent that auto-reviews plans and complex changes using a complementary model (experimental for Claude models), plus better session grouping, a permission bypass when you explicitly allow a tool, and a new async notification hook for completions and prompts 1.0.18 notes. Last release bundled built-in skills, sped up session resume, and added an HTTPS OAuth fallback for MCP flows (helps with providers like Slack) releases.
Agent workflows are stabilizing across IDE and CLI, making it easier to standardize how developers plan, edit, and review code.
The new CLI Critic and hooks let teams automate early review and wire agent events into chat or observability.
-
terminal
Install copilot-cli v1.0.18, enable the experimental Critic for Claude models, and run it against a real migration or service refactor plan.
-
terminal
Update VS Code and Copilot to versions that include agent skills, then trial a simple custom skill and measure prompt fatigue with the new permission bypass.
Legacy codebase integration strategies...
- 01.
Use the preToolUse permissionDecision 'allow' to cut repetitive prompts in existing CLI workflows and pipe the new notification hook to Slack.
- 02.
Validate MCP OAuth HTTPS fallback with your IdP/Slack; keep BYOK in preview in mind before routing regulated data through Copilot.
Fresh architecture paradigms...
- 01.
Design an agent-first inner loop: VS Code agent skills for editing, Copilot CLI for planning/critique, and MCP for tool integrations.
- 02.
Codify reusable skills and custom agents early to make repo- or domain-specific context a first-class capability.