What Happened
GitHub’s October 5 Copilot release made Claude Haiku 5.5 available to Pro, Pro+, Max, Business, and Enterprise users. Local sandboxing is now generally available at no extra cost in Copilot CLI, the Copilot app, and VS Code Agent Host sessions. It limits agent access to files, networks, and credentials. The Copilot app also supports using separate GitHub accounts for a Copilot license and repository access, while the CLI can discover models from a running local Ollama instance through /model. VS Code 1.141 adds side-by-side agent-session comparison and cleanup of inactive session worktrees. [2]
In JetBrains, enterprise administrators can set the default model for new Copilot agent conversations, without removing users’ explicit model choice. A setting can prevent automatic MCP server startup for Copilot and Claude. Chat and diagnostics gain navigation, message editing and copying, and an inline-chat Fix action. The release also addresses reliability across agent sessions, model switching, MCP configuration, customizations, and worktrees. The plugin now requires JetBrains IDE 2025.2 or later. [1]
These notes establish changes for GitHub Copilot, VS Code, and JetBrains integrations; they do not establish new releases for Cursor, Windsurf, Replit, Sourcegraph, or Continue.dev.
Why It Matters to Businesses
Agent permissions are becoming a deployment decision, not just a developer preference. Sandboxing provides a way to constrain local execution, while the JetBrains MCP setting lets teams decide whether external tools start automatically. Neither replaces repository permissions or credential controls. [1][2]
Model defaults can make an enterprise rollout more consistent, but users can still select another model in JetBrains. Teams that need strict model restrictions should verify whether their existing licensing and policy controls meet that requirement rather than treating a default as enforcement. [1]
Kimbodo Engineering Perspective
We would prioritize permission boundaries and reproducible workflows over enabling every new agent feature at once. Side-by-side sessions can help review competing changes; worktree cleanup reduces disk use but should happen only after changes are committed, discarded, or otherwise preserved. Separate license and repository accounts are useful for organizations with distinct identity arrangements, provided access is tested against the intended repositories. [2]
How We Would Implement It
- Pilot Copilot CLI, the app, and VS Code Agent Host with local sandboxing enabled. Test permitted file paths, network destinations, and credential access against representative development tasks. [2]
- Inventory MCP servers and their permissions; disable automatic startup where a server is not needed for the task. [1]
- Set a JetBrains default model, document when developers should choose another, and confirm participating IDEs are on version 2025.2 or later. [1]
- Define a session workflow: compare agent changes, run tests and human review, then clean up inactive worktrees. [2]
- Validate separate Copilot-license and repository accounts with least-privilege access before a broader rollout. [2]
Risks, Costs and Security
Local sandboxing is available at no extra cost, but rollout still takes engineering time for policy testing, support, and review. A sandbox limits specified access; it does not make generated code correct or prevent a user from approving an unsafe change. MCP servers and local Ollama models add configuration paths that should be inventoried and tested before use with sensitive code. Finally, the JetBrains minimum-version change may require IDE upgrades before affected teams can adopt the plugin release. [1][2]
Where Kimbodo Comes In
Kimbodo builds and operates this in production for businesses — see our AI Application Development practice, or Estimate My AI Application.
