session: auto-continue after a stream interruption is delivered as an unmarked synthetic user message #56295

Activity

1. opencode-agent commented on Oct 17, 2026

   

   opencode-agentBot

   on Oct 17, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the detailed report. I reproduced this on 2.0.26, and it still happens on the latest dev (512165c).

   Setup: the simulated OpenAI Chat model streams "Plan: push the fix and comment on the issue. Reply GO to approve." and then the connection drops before any finish_reason. OpenCode retries, and the next request to the model ends with:

   user:      Fix the bug, but wait for my GO before pushing.
   assistant: Plan: push the fix and comment on the issue. Reply GO to approve.
   user:      The previous response was interrupted. Continue from where you left off without repeating completed content.
   

   The continuation goes out as a plain user message with nothing marking it as automatic. Right after an approval request, it looks like the user replying. All synthetic messages are converted this way in packages/core/src/session/runner/to-llm-message.ts, while system messages already go through the <system-update> / developer path.

   One limit: I cut the stream with a dropped connection rather than your exact "stream ended without finish_reason" error. Both lead to the same continuation code. The simulated model also can't show a real model taking this as approval; the test only shows that nothing marks the message as automatic.

   The recording and screenshot below show the run, and the script below reproduces it.

   recording.mp4

   

   Reproduction script (drive.ts)

   // Repro for anomalyco/opencode#56295: after an OpenAI Chat stream ends without
   // finish_reason, OpenCode auto-continues by sending a plain user-role message,
   // indistinguishable from real user input.
   import { Effect, Stream } from "effect"
   import { Llm, OpenCodeDriver } from "opencode-drive"
   
   const requests: Array<unknown> = []
   
   export default OpenCodeDriver.use(
     {
       ...(process.env.OC_DEV ? { opencode: { dev: process.env.OC_DEV } } : {}),
       tui: { recording: true },
     },
     ({ ui, llm }) =>
       Effect.gen(function* () {
         yield* llm.serve((request: any, index) => {
           requests.push(request.body)
           if (index === 0)
             // Ask for approval, then cut the stream before any finish_reason arrives.
             return Stream.make(
               Llm.text("Plan: push the fix and comment on the issue. Reply GO to approve."),
               Llm.disconnect(),
             )
           return Stream.make(Llm.text("Continuing."))
         })
         yield* ui.submit("Fix the bug, but wait for my GO before pushing.")
         yield* ui.waitFor("Continuing.", { timeout: 30000 }).pipe(
           Effect.catchTag("UiWaitTimeoutError", (e: any) =>
             Effect.sync(() => console.log("TIMEOUT; requests:", requests.length, JSON.stringify(e.frame?.lines?.map((l: any) => l.text ?? JSON.stringify(l)).join("\n")).slice(0, 4000))),
           ),
         )
         yield* ui.screenshot("after-continue")
         const last: any = requests[requests.length - 1]
         const msgs = last?.messages ?? []
         console.log("REQUEST COUNT", requests.length)
         console.log("LAST REQUEST MESSAGES (tail):")
         console.log(JSON.stringify(msgs.slice(-3), null, 2))
       }),
   )

   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.
2. opencode-agent

   added

   reproduced

   severity:high

   on Oct 17, 2026
3. rekram1-node

   closed this as completedin #60674on Nov 10, 2026