Copilot CLI Gains Cross-Session Memory
GitHub Copilot's command-line interface has been updated with new agentic workflow capabilities. The tool now features cross-session memory, allowing it to build project-level context over time rather than just providing per-file suggestions. It also now supports remote plugins loaded from any GitHub repository, enabling developers to script and extend Copilot's behavior.
- The cross-session memory is an experimental feature shipped in version v0.0.412 that enables the CLI to recall context from previous sessions, such as files edited and pull requests opened. This addresses a common issue with AI assistants where each session starts without knowledge of prior work. - GitHub's implementation of memory verifies information in real-time by storing each memory with citations to specific code locations. When a memory is retrieved, it checks the citations against the current codebase to avoid acting on stale information. - The introduction of remote plugins in version v0.0.413-0 allows developers to load plugins from any Git repository, not just those bundled in the official marketplace. This change was implemented shortly after the community raised the need for it. - These new capabilities are part of a broader push by GitHub to create "agentic workflows" that can automate more complex development tasks like triaging issues and continuous documentation. These workflows can be executed by different AI engines, including Copilot CLI, Claude Code, or OpenAI Codex. - To manage the context window, the Copilot CLI has features like automatic and manual compaction of history, as well as a command to visualize token usage. It also allows users to resume previous local and remote sessions to continue with existing context. - The Copilot CLI now includes specialized built-in agents for common tasks such as exploring a codebase, running tests, creating implementation plans, and reviewing code. - Early A/B tests on a related cross-agent memory system for Copilot showed a 7% increase in pull request merge rates and a 2% increase in positive feedback for code review comments. - The CLI complements the in-editor Copilot extension by focusing on terminal-based workflows. For example, a developer might use the IDE extension to generate a Dockerfile and then use the CLI to build the image and containerize the application.