tui: reverting during generation restores the prompt but does not revert #57383

Activity

1. opencode-agent commented on Oct 20, 2026

   

   opencode-agentBot

   on Oct 20, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the clear report. I could reproduce this on 2.1.0, and it still happens on the latest dev (d814879).

   What I did, using a simulated model that streams a long reply:

   1. Sent a prompt and, while the reply was still streaming, opened the user message's Message Actions (through /timeline).
   2. Chose Revert.

   The prompt was filled with the old message immediately, and a Session is busy: ses_… toast appeared. The turn was not interrupted and streamed to the end. The original message and the reply stayed in the transcript, so nothing was reverted, and the old text was left in the prompt, ready to be sent again.

   The Revert action fills the prompt before it attempts the revert and never interrupts the running turn, unlike /undo, which interrupts, waits, stages and only then fills the prompt. The recording and screenshots below show it, and the script below reproduces it.

   Related: #53159 (same symptom on v1; its fix PR #53161 was not merged), #14014 and #43670 (the same problem in the Web UI).

   recording.mp4

   

   

   Reproduction script (drive.ts)

   // Repro for anomalyco/opencode#57383: Message Actions -> Revert while a turn is
   // streaming fills the prompt, but the revert fails (session busy) and the turn keeps running.
   import { Effect } from "effect"
   import { Llm, OpenCodeDriver } from "opencode-drive"
   
   const long = Array.from({ length: 40 }, (_, i) => `Chunk${i + 1}`).join("\n\n")
   
   export default OpenCodeDriver.use(
     {
       ...(process.env.OC_DEV ? { opencode: { dev: process.env.OC_DEV } } : {}),
       project: { git: true, files: { "src/value.ts": "export const value = 1\n" } },
       tui: { recording: true, keypressOverlay: true },
     },
     ({ ui, llm }) =>
       Effect.gen(function* () {
         // A slow streaming response so the turn is still running when we revert.
         yield* llm.queue(Llm.text(long, { delay: 300, chunkSize: 5 }))
         yield* ui.submit("Write a long answer about value.ts")
         yield* Effect.sleep(5000)
         yield* ui.screenshot("01-streaming")
   
         // Open the timeline and choose the user message -> Message Actions
         yield* ui.type("/timeline")
         yield* Effect.sleep(500)
         yield* ui.enter()
         yield* Effect.sleep(1000)
         yield* Effect.sleep(500)
         yield* ui.screenshot("02-timeline")
         yield* ui.enter()
         yield* ui.waitFor("Message Actions", { timeout: 15000 })
         yield* ui.screenshot("03-message-actions")
   
         // Pick "Revert"
         yield* ui.type("Revert")
         yield* Effect.sleep(300)
         yield* ui.enter()
         yield* Effect.sleep(1500)
         yield* ui.screenshot("04-after-revert")
   
         // The turn keeps streaming and the old message is still there.
         yield* Effect.sleep(4000)
         yield* ui.screenshot("05-still-running")
         // Let the turn finish: it was never interrupted, and nothing was reverted.
         yield* Effect.sleep(15000)
         yield* Effect.sleep(1500)
         yield* ui.screenshot("06-turn-finished-not-reverted")
       }),
   )

   Run it with opencode-drive: opencode-drive run ./drive.ts. It drives opencode from your PATH; set OC_DEV to the path of an OpenCode checkout to run that from source instead.

   🚀React with 🚀1Reacted by Yassine Safraoui
2. opencode-agent

   added

   reproduced

   on Oct 20, 2026
3. yassine-safraoui commented on Oct 20, 2026

   

   yassine-safraoui

   on Oct 20, 2026

   Author

   More actions

   Omg you guys really cooked with the opencode-agent 😵