Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Python agentagent-osaiai-agentsai-partnerchatopsclaudeclaude-codecodexdevtoolsdiscord-botlark-botllmlocal-firstlong-term-memoryopencodeslack-botvibe-codingvibeywechat
Refactor Plan: Unify OpenCode Poll Loop
Current Architecture (Problem)
Two separate polling loops with duplicated logic:
Main Poll Loop (
_process_message, line 1671-2223)- Polls for new messages after initial prompt
- Detects question → sends buttons → exits function
- Lives in its own function call stack
Post-Answer Poll Loop (
_process_question_answer, line 1323-1454)- User clicks button → new HTTP request → new function call
- Submits answer → polls for completion
- Duplicates almost all logic from main loop
Issues
- Code Duplication: 90% identical polling logic in two places
- Inconsistent Behavior: Originally had timeout in post-answer but not main
- Maintenance Burden: Bug fixes need to be applied twice
- Conceptual Complexity: Why two loops for the same task?
Target Architecture (Solution)
Single unified polling loop that handles questions internally:
async def _process_message(request):
await server.prompt_async(...)
# Single unified loop
while True:
messages = await server.list_messages(...)
for message in messages:
# Process tool calls
for part in message.parts:
if part.tool == "question":
if part.state.status != "completed":
# New question detected
await self._send_question_to_user(...)
# Wait for answer (blocks until user responds)
await self._wait_for_answer_event(session_id)
# Answer submitted, continue loop
# If status == "completed", answer already submitted, continue
else:
# Other tool calls
await send_tool_call(...)
# Emit intermediate messages
if message.finish == "tool-calls":
await send_message(...)
# Check completion
if last_message.finish != "tool-calls":
break
await asyncio.sleep(2.0)
# Send final result
await send_final_result(...)
async def _process_question_answer(request, pending):
# Simplified: only submits answer
await server.reply_question(...)
# Signal main loop to continue
event = self._question_answer_events.get(session_id)
if event:
event.set()
Implementation Steps
Step 1: Add event coordination ✅ COMPLETE
- Add
_question_answer_events: Dict[str, asyncio.Event]to class - Create event when question detected (
_get_or_create_question_event) - Set event when answer submitted
- Clear event after processing (
_clear_question_event) - Add timeout tracking set (
_timed_out_questions) - Commit:
72c3c99- Add question answer event coordination
Step 2: Refactor main loop to not exit on question ✅ COMPLETE
- Remove
returnafter sending question buttons - Add
await event.wait()to block until answer (_wait_for_question_answer) - Continue processing after answer submitted (restart poll loop)
- Add timeout handling (30 minutes)
- Commit:
2c1e928- Unify poll loop - wait for answer instead of exit
Step 3: Simplify _process_question_answer ✅ COMPLETE
- Remove entire post-answer polling loop (~150 lines deleted)
- Keep only answer submission logic
- Set event to resume main loop
- Update routing to not cancel poll task
- Commit:
2806bd2- Simplify question answer handler and update routing
Step 4: Handle edge cases ✅ COMPLETE
- Timeout for waiting on answer (user never responds) - 30 minute timeout
- Late answer race condition -
_timed_out_questionsset prevents resumption - Answer submission failures - set event even on error to unblock loop
- Immediate poll restart - exit message loop when
restart_poll=True - Safety wrapper for timeout handler - try-except protection
- Nested questions (question after question) - needs testing
- Concurrent requests to same session - needs testing
- Poll restoration on restart - needs testing
- Commits:
39cbb72,517886f- Fix timeout and P0 issues
Step 5: Testing 🚧 IN PROGRESS
- Test normal question flow
- Test nested questions
- Test timeout scenarios (reduce timeout to 30s for testing)
- Test answer submission failures
- Test late answer after timeout
- Test concurrent operations
- Test poll restoration
- Test
/stopduring question wait
Benefits
- Single Source of Truth: One polling loop, one place to fix bugs
- Consistent Behavior: Same timeout/error handling everywhere
- Simpler Mental Model: One continuous flow instead of disconnected loops
- Easier Debugging: Single execution path through the code
- Better Maintainability: Changes only need to be made once
Risks
- Breaking Changes: Need thorough testing
- Complexity: Event-based coordination might be tricky
- Edge Cases: Concurrent questions, restarts, etc.
Current Status
Branch: refactor/unify-opencode-poll-loop
Base Branch: fix/opencode-question-poll-resume (PR #28 - awaiting merge)
Completed:
- ✅ All core refactoring (Steps 1-3)
- ✅ Critical edge cases handled (Step 4)
- ✅ Syntax validation passed
Next Steps:
- Testing Phase - Validate all scenarios work correctly
- Decide Merge Strategy:
- Option A: Merge PR #28 first, then merge refactor as separate PR
- Option B: Update refactor to include all fixes from PR #28, merge directly
- Production Deployment - Monitor for issues
Testing Checklist
Critical Scenarios
- Normal flow: Send prompt → question appears → answer → completion
- Timeout: User doesn't answer for 30 minutes (reduce to 30s for testing)
- Late answer: User clicks button after timeout
- Answer failure: Simulate
reply_question()exception - Nested questions: OpenCode asks second question after first answer
- Cancellation: User sends
/stopwhile waiting for answer - Empty response: Final message has no text content (original bug)
Testing Commands
# Reduce timeout for faster testing (in opencode_agent.py)
QUESTION_WAIT_TIMEOUT_SECONDS = 30 # Change from 30*60
# Start vibe in editable mode
# Run from the Avibe repository root.
uv tool install --force --editable .
vibe
# Test with a task that asks questions
# Example: "Search Twitter for AI news and ask me which results to analyze"
Migration Strategy
- ✅ Implement refactor on feature branch
- 🚧 Run extensive manual testing
- Create PR and request code review
- Deploy to staging/production
- Monitor logs for issues:
~/.vibe_remote/logs/vibe_remote.log