[BUG] UnknownError: prompt_async failed — race condition between session cancel and Todo Continuation / Question dismiss #20250

Bug Description

UnknownError: prompt_async failed occurs due to a race condition between session lifecycle (cancel/idle) and automatic prompt submission. This happens in two scenarios:

1. Todo Continuation: AI completes all agentic loop steps → session exits loop and gets cancelled → TUI detects pending todos → auto-triggers prompt_async → session is already in cancelled/idle state → UnknownError
2. Question/ESC dismiss: AI presents a Question tool with options → user presses ESC to dismiss → abort fires → session goes idle → an immediate prompt_async is sent → UnknownError

Reproduction

Scenario 1: Todo Continuation (most common)

1. Start a session with a work plan that has pending TODO items
2. Let the AI agent complete its agentic loop (step N → "exiting loop")
3. Session gets cancelled and goes idle
4. TUI detects remaining todos and auto-fires prompt_async (Todo Continuation)
5. prompt_async fails because session is already in cancelled/idle state
6. Error: UnknownError prompt_async failed

Scenario 2: ESC on Question

1. AI uses the Question tool to present selectable options to the user
2. User presses ESC to dismiss the question (instead of selecting an option)
3. TUI sends POST /session/{id}/abort
4. Session transitions to cancelled/idle
5. Immediately after, a prompt_async is fired (likely from queued/resumed operation)
6. Error: UnknownError prompt_async failed

Logs

Scenario 1 — Todo Continuation (from log 2026-03-30T103243.log, session ses_2c1595b7cffeNCN7U4nFQfGGda)

12:23:00  step=22 loop          ← AI completes final agentic step
12:23:01  exiting loop           ← Loop exits normally
12:23:01  compaction pruning
12:23:01  session.prompt cancel  ← Session marked as cancelled
12:23:01  session.idle           ← Session enters idle state
12:23:01  command.executed       ← Command completes

12:23:01  GET  /session/.../todo     ← TUI fetches todo list (finds pending items)
12:23:01  GET  /session/.../message
12:23:01  POST /session/.../prompt_async   ← TUI auto-triggers Todo Continuation
12:23:01  ERROR UnknownError prompt_async failed   ← ❌ Session is already cancelled!

12:23:13  step=0 loop           ← Second continuation attempt starts
12:23:27  step=1 loop
12:23:28  exiting loop          ← Exits again
12:23:28  POST /session/.../prompt_async
12:23:28  ERROR UnknownError prompt_async failed   ← ❌ Same error again!

This happened twice in the same session within ~27 seconds.

Scenario 2 — ESC on Question (from log 2026-03-30T124351.log, session ses_2c13a106effeokLMx959pcfGgf)

12:58:16  message.part.delta (continuous)   ← AI streaming output with Question tool
          ... (user inactivity, Question displayed) ...

13:10:54  POST /session/.../abort           ← User pressed ESC
13:10:54  session.prompt cancel             ← Cancel signal
13:10:54  session.idle                      ← Goes idle
13:10:54  abort completed (3ms)

13:10:54  POST /session/.../prompt_async    ← Immediate prompt_async
13:10:54  ERROR UnknownError prompt_async failed   ← ❌ Race condition!
13:10:54  ERROR Aborted process

This happened 3 times in the same session (13:10:54, 13:11:00, 13:12:05).

Expected Behavior

• Todo Continuation: When the session is in cancelled/idle state, Todo Continuation should either skip the auto-prompt, or re-initialize the session properly before sending prompt_async.
• ESC on Question: Dismissing a Question should not cascade into a prompt_async error. The abort should consume any queued prompt operations.

Environment

• OpenCode: v1.3.x (oh-my-opencode 3.14.0)
• OS: Linux
• Backend: Running with multiple LLM providers

Root Cause Analysis

The core issue is a state lifecycle race condition:

1. Session transitions: active → cancelling → idle
2. prompt_async is dispatched after cancel but before the session is ready to accept new prompts
3. No guard exists to check session state before dispatching prompt_async

Suggested Fix

Add a session state check in prompt_async handler:

// Before executing prompt_async, verify session is in a valid state
if (session.state === 'cancelled' || session.state === 'idle') {
  // Option A: Re-initialize session then proceed
  await session.reinitialize();
  // Option B: Silently ignore and don't fire
  return;
}

Alternatively, the Todo Continuation mechanism should wait for the session to fully settle (post-cancel cleanup) before auto-firing.

Workaround

Run /stop-continuation to disable Todo Continuation before the AI completes its loop. For ESC dismissals, no workaround currently exists (the error is harmless but noisy).