GitHub Copilot CLI adds prompt scheduling, voice input, and rubber duck support, expanding terminal agents beyond one-shot commands. Read now.
Terminal agents move past one-shot prompts
Most developer CLI tools still treat the agent like a single request: you type a prompt, wait for a reply, then decide what to do next. That pattern works for quick refactors and one-off questions, but it breaks down when the work spans idle time, recurring checks, or a longer thinking loop. GitHub Copilot CLI’s push into prompt scheduling, voice input, and rubber duck support aims at that gap—keeping the agent useful in the terminal even when you are not sitting in a tight command-and-response cycle.
The shared idea is continuity. Instead of every interaction starting from a blank line, you can queue work, speak without breaking flow, and use the agent as a sounding board when the hard part is clarifying the problem rather than generating code immediately.
Prompt scheduling for deferred and recurring work
Scheduling turns a prompt into something you can run later or on a cadence, not only the moment you hit enter. That fits work you already do around the terminal: “check this after the build finishes,” “re-run the investigation when I am free,” or “nudge me with a review of this module on a regular interval.” The agent stops being only an interrupt-driven helper and becomes a lightweight job runner for developer intent.
Use scheduling when the cost of context-switching is high. Draft the prompt carefully once—include the repo path, the goal, constraints, and what “done” looks like—then let the run happen when you are not mid-flow. Review scheduled output the same way you review any agent result: verify commands, inspect diffs, and treat suggestions as proposals rather than automatic merges. Keep schedules narrow so a recurring job does not spam noise or re-open work that is already settled.
Voice input without leaving the shell
Voice input reduces friction when typing is awkward: hands on a second machine, walking through a whiteboard, or describing a bug you just reproduced. Speaking a prompt can be faster for narrative context—“here is the error, here is what I already tried”—than compressing the same story into a terse one-liner.
Good voice prompts still need structure. State the goal first, then constraints, then files or commands that matter. After transcription, skim the text before you submit so misheard names and paths do not send the agent down the wrong branch. Voice is strongest as a capture tool for intent; precision still belongs in a quick edit pass before the agent runs.
Rubber duck mode as deliberate reasoning support
Rubber ducking is the practice of explaining a problem out loud so gaps in your assumptions become obvious. Rubber duck support in Copilot CLI formalizes that loop: you walk through symptoms, hypotheses, and failed attempts; the agent asks clarifying questions, restates the problem, and helps separate root cause from side effects before anyone writes a large patch.
- Use it early when the bug is vague or multi-service and you need a crisp problem statement.
- Feed it concrete evidence—error text, stack traces, recent changes—not only gut feel.
- End each session with an agreed next experiment so talk turns into a single, testable step.
Together, scheduling, voice, and rubber duck mode expand terminal agents beyond one-shot commands into timed work, low-friction capture, and structured thinking. The practical win is not more automation for its own sake; it is keeping investigation and follow-through inside the same place you already build and debug.