AddThisFeature

Stop and Regenerate Controls

Let users halt a bad answer and ask again without losing the conversation.

moderate AI Assistants

What it adds

Stop and regenerate actions on an in-progress or completed AI response, with the prior context preserved.

What your agent is told to do

5
  1. 1

    Wire stop to an actual cancellation of the server-side run, propagated to the provider request, so generation ends rather than continuing invisibly.

  2. 2

    Keep whatever text arrived before the stop, store it marked as stopped by the user, and make that state visible in the transcript rather than deleting it.

  3. 3

    Build regenerate from the conversation state as it stood before the response being replaced, never from the abandoned partial output.

  4. 4

    Retain previous attempts and let the user move between them, so regenerating does not destroy an answer that was better than its replacement.

  5. 5

    Streaming AI Responses owns transport and incremental rendering. This entry owns the user-facing controls and the run lifecycle around them. Do not reimplement stream handling here.

Edge cases it handles

8
  • Hiding incoming text without aborting the request leaves the model generating and the account paying. Stop must cancel the run at the server and at the provider where the provider supports it.
  • Partial output that is silently discarded looks like a bug, and partial output shown as if complete is worse. Keep it, label it as stopped, and disable any action that treats it as a finished answer.
  • Regenerating from the unfinished response feeds the model its own truncated text and produces a continuation rather than a fresh answer. Rebuild the request from the ancestor messages only.
  • Usage accounting must reflect what actually happened: a stopped run consumes what it generated, and a regeneration is a new run, not a free retry. Reconcile against the provider's reported usage rather than assuming.
  • A tool call that already wrote data, sent a message, or charged something cannot be undone by stopping. Detect completed side effects, surface them to the user, and never replay them on regenerate.
  • Rapid repeated presses of regenerate must not launch parallel runs against the same message. Disable the control while a run is in flight.
  • Stopping must leave the conversation in a state the user can continue from, with the input available and no orphaned pending row in the transcript.
  • A run that has already finished server-side when stop is pressed should complete cleanly rather than being recorded as stopped.

Definition of done

8
  • Stop cancels the run at the server and the provider, and generation verifiably ends.
  • Partial output is retained in the transcript, clearly marked as stopped, and cannot be mistaken for a complete answer.
  • Regeneration rebuilds the request from prior context and never includes the abandoned partial response.
  • Stopped runs and regenerations are accounted for accurately against real reported usage.
  • Tool calls with completed side effects are surfaced and never re-executed by a regeneration.
  • Repeated presses cannot start parallel runs, and previous attempts remain reachable.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.