GitHub Copilot for JetBrains 1.18 adds AI‑assisted approvals, the ability to rewrite prior agent messages, organization‑wide shared skills, and persistent controls for the built‑in MCP server. For engineers who embed AI‑driven code generation into CI/CD pipelines or on‑prem IDEs, the update promises fewer manual interruptions while exposing new policy‑enforcement knobs that must be tracked.
AI‑assisted approvals reduce low‑risk interruptions
The plugin now runs a public preview called assisted approvals. When a Copilot agent requests a low‑risk tool call, the system auto‑approves it; higher‑risk actions still trigger an explicit prompt. This split‑path reduces the number of pop‑ups for routine operations such as formatting or simple refactors, while preserving human oversight for potentially disruptive changes like dependency upgrades or secret handling.
Re‑editing prior messages improves session flow
Users can now replace an earlier request in an active agent session. The plugin rewinds both the conversation history and any file modifications before applying the new input, so the session continues from the corrected point instead of appending a corrective message. Practically, this means fewer stray edits and a cleaner audit trail when reviewing what the agent did.
Shared skills and organization‑managed instructions
Both local and agent sessions now accept organization and enterprise skill bundles, plus custom instructions defined at the org level. Teams can distribute standard prompts, coding conventions, or security checks across all developers without configuring each IDE individually. The feature introduces a governance layer that can be audited for compliance, but it also requires coordination to keep the shared skill set in sync with evolving policies.
Persistent MCP tool controls give operators more visibility
A new toggle lets administrators enable or disable the built‑in GitHub MCP Server without touching manually configured servers. In addition, each MCP tool now has a persistent per‑tool control, allowing operators to whitelist or blacklist specific integrations while the built‑in server remains on by default. This change centralises tool management, but it also adds a configuration surface that must be reviewed during platform hardening.
UI and reliability tweaks
The session toolbar now includes a side‑by‑side chat panel switcher, letting developers keep a conversation open while browsing other sessions. Usage tips, a simplified welcome screen, and restored shortcuts improve discoverability. Under the hood, inline chat reliability has been hardened, edit preservation during request failures is fixed, and compatibility with IntelliJ 2026.3 EAP builds is restored. Inline chat is hidden in JetBrains Gateway and remote development environments, and users on IDE version 2025.1 receive an upgrade notice.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Adopt the new version if you rely on Copilot for routine code generation, but audit the assisted‑approval thresholds to ensure they align with your risk model. Propagate shared skills through your org’s instruction repository and treat them as part of your configuration baseline. Review the persistent MCP toggles and per‑tool settings during your next security hardening cycle, and verify that any legacy IDE installations are upgraded past 2025.1 to avoid deprecation warnings. Finally, incorporate the re‑edit capability into your agent‑driven workflow to keep change histories clean and reversible.
