Copilot CLI just got a new built-in subagent, Rubber Duck 🐤 It pulls in a different LLM (GPT <-> Claude) to critique your work, verify plans and keep sessions on track. Auto-invoked at key points or run it yourself (“critique” / “run Rubber Duck”). Enable with "/experimental on"
@ssk_bl Putting this into your instruction file (copilot-instructitons.md, AGENTS.md, etc.) should achieve that: "When using plan mode, store your plan.md in the current session working directory instead of SESSION_DIR"
Copilot CLI just got a nice QoL improvement: you can export your session timeline with "/share html" as a self-contained HTML file with built-in search and filtering. I’ve found this especially useful for visualizing what happened during a session and debugging agent workflows.
Recently we shipped extensions in Copilot CLI, letting you extends the agent's core capabilities. To experiment, I wrote one that pairs your Telegram bot with a CLI session so you can control it from your phone. Code with install & setup instruction at https://t.co/KYSwmT48hv